Datasets:
engine large_string | model_id large_string | lang large_string | utt_id large_string | duration_sec float64 | ref_text_raw large_string | ref_text_normalized large_string | hypothesis large_string | latency_sec float64 | timestamp_utc large_string | retries int64 | error null | params_json large_string | raw_response_json large_string |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
amivoice | -a-general | ja | ja_0000 | 10.44 | インターネットで「敵対的環境コース」について検索すると、おそらく現地企業の住所が出てくるでしょう。 | インターネットで 敵対的環境コース について検索すると おそらく現地企業の住所が出てくるでしょう | インターネットで敵対的環境コースについて検索すると、おそらく現地企業の住所が出てくるでしょう。 | 1.9323 | 2026-07-04T13:29:42Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.97800004, "endtime": 8492, "rulename": "", "starttime": 0, "tags": [], "text": "インターネットで敵対的環境コースについて検索すると、おそらく現地企業の住所が出てくるでしょう。", "tokens": [{"confidence": 1.0, "endtime": 2524, "spoken": "いんたーねっと", "starttime": 1804, "written": "インターネット"}, {"confidence": 1.0, "e... |
amivoice | -a-general | ja | ja_0001 | 14.22 | また、北側に行くなら世界的に有名なマリア像の聖地であるファティマの聖母の聖域(神社)を訪れましょう。 | また 北側に行くなら世界的に有名なマリア像の聖地であるファティマの聖母の聖域神社を訪れましょう | また、北側に行くなら、世界的に有名なまりあ像の聖地であるファティマの聖母の聖域神社を訪れましょう。 | 2.399 | 2026-07-04T13:29:45Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.981, "endtime": 11484, "rulename": "", "starttime": 3100, "tags": [], "text": "また、北側に行くなら、世界的に有名なまりあ像の聖地であるファティマの聖母の聖域神社を訪れましょう。", "tokens": [{"confidence": 1.0, "endtime": 3820, "spoken": "また", "starttime": 3420, "written": "また"}, {"confidence": 0.71, "endtime":... |
amivoice | -a-general | ja | ja_0002 | 12.84 | バルセロナの公用語はカタルーニャ語とスペイン語です。約半数がカタルーニャ語を好み、大多数がカタルーニャ語を理解し、ほぼ全員がスペイン語を知っています。 | バルセロナの公用語はカタルーニャ語とスペイン語です 約半数がカタルーニャ語を好み 大多数がカタルーニャ語を理解し ほぼ全員がスペイン語を知っています | バルセロナの公用語はカタルーニャ語とスペイン語です。約半数が語る映画を好み、大多数が語る旨ご理解し、ほぼ全員がスペイン語を知っています。 | 2.4638 | 2026-07-04T13:29:48Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.883, "endtime": 10982, "rulename": "", "starttime": 1750, "tags": [], "text": "バルセロナの公用語はカタルーニャ語とスペイン語です。約半数が語る映画を好み、大多数が語る旨ご理解し、ほぼ全員がスペイン語を知っています。", "tokens": [{"confidence": 0.69, "endtime": 2646, "spoken": "ばるせろな", "starttime": 2022, "written": "バルセロナ"}, {"con... |
amivoice | -a-general | ja | ja_0003 | 12.54 | その長い顎には70本以上の鋭い歯が並び、口蓋には別の歯列があり、つまりここを通ったら逃げ道はないということになります。 | その長い顎には70本以上の鋭い歯が並び 口蓋には別の歯列があり つまりここを通ったら逃げ道はないということになります | その長い顎には70本以上の鋭い歯が並び、郊外には別の私立があり、つまりここを通ったら逃げ道はないということになります。 | 2.4189 | 2026-07-04T13:29:51Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.964, "endtime": 10292, "rulename": "", "starttime": 1700, "tags": [], "text": "その長い顎には70本以上の鋭い歯が並び、郊外には別の私立があり、つまりここを通ったら逃げ道はないということになります。", "tokens": [{"confidence": 1.0, "endtime": 2372, "spoken": "その", "starttime": 2036, "written": "その"}, {"confidence": 0.83, ... |
amivoice | -a-general | ja | ja_0004 | 12.06 | ロスビー数が小さいほど磁気反転に関して星の活性が低下するわけです。 | ロスビー数が小さいほど磁気反転に関して星の活性が低下するわけです | ロスB数が小さいほど、磁気判定に関して星の活性が低下するわけです。 | 2.0157 | 2026-07-04T13:29:54Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.995, "endtime": 8994, "rulename": "", "starttime": 3250, "tags": [], "text": "ロスB数が小さいほど、磁気判定に関して星の活性が低下するわけです。", "tokens": [{"confidence": 0.66, "endtime": 3906, "spoken": "ろす", "starttime": 3570, "written": "ロス"}, {"confidence": 0.58, "endtime": 4098, "spoken":... |
amivoice | -a-general | ja | ja_0005 | 6.96 | 群島や湖では、必ずしもヨットは必要ありません。 | 群島や湖では 必ずしもヨットは必要ありません | 群島や湖では必ずしもヨットは必要ありません。 | 1.5586 | 2026-07-04T13:29:56Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.997, "endtime": 6732, "rulename": "", "starttime": 1100, "tags": [], "text": "群島や湖では必ずしもヨットは必要ありません。", "tokens": [{"confidence": 0.87, "endtime": 2108, "spoken": "ぐんとう", "starttime": 1340, "written": "群島"}, {"confidence": 0.99, "endtime": 2444, "spoken": "や", "st... |
amivoice | -a-general | ja | ja_0006 | 8.88 | パリジャンは、自己中心的で横柄で失礼な人が多いと言われています。 | パリジャンは 自己中心的で横柄で失礼な人が多いと言われています | パリジャンは自己中心的な大柄で失礼な人が多いと言われています。 | 2.2171 | 2026-07-04T13:29:59Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.98099995, "endtime": 6890, "rulename": "", "starttime": 0, "tags": [], "text": "パリジャンは自己中心的な大柄で失礼な人が多いと言われています。", "tokens": [{"confidence": 0.97, "endtime": 2266, "spoken": "ぱりじゃん", "starttime": 1498, "written": "パリジャン"}, {"confidence": 1.0, "endtime": 2426, "spo... |
amivoice | -a-general | ja | ja_0007 | 12.96 | メインステージの音楽が終わっても、フェスティバルには夜遅くまで演奏を流し続けるセクションがあるかもしれないことを覚えておいてください。 | メインステージの音楽が終わっても フェスティバルには夜遅くまで演奏を流し続けるセクションがあるかもしれないことを覚えておいてください | メインステージの音楽が終わっても、フェスティバルには喜則まで演奏を流し続けるセクションがあるかもしれないことを覚えておいてください。 | 2.0096 | 2026-07-04T13:30:02Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.993, "endtime": 10842, "rulename": "", "starttime": 1850, "tags": [], "text": "メインステージの音楽が終わっても、フェスティバルには喜則まで演奏を流し続けるセクションがあるかもしれないことを覚えておいてください。", "tokens": [{"confidence": 1.0, "endtime": 2954, "spoken": "めいんすてーじ", "starttime": 2154, "written": "メインステージ"}, {"co... |
amivoice | -a-general | ja | ja_0008 | 9.48 | 参列者の数が多すぎて、全員がサンピエトロ広場での葬儀に参加することは不可能でした。 | 参列者の数が多すぎて 全員がサンピエトロ広場での葬儀に参加することは不可能でした | 参列者の数が多すぎて、全員がサンピエトロ広場での葬儀に参加することは不可能でした。 | 1.553 | 2026-07-04T13:30:04Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.972, "endtime": 7790, "rulename": "", "starttime": 1150, "tags": [], "text": "参列者の数が多すぎて、全員がサンピエトロ広場での葬儀に参加することは不可能でした。", "tokens": [{"confidence": 1.0, "endtime": 2030, "spoken": "さんれつ", "starttime": 1454, "written": "参列"}, {"confidence": 0.99, "endtime": 2174, ... |
amivoice | -a-general | ja | ja_0009 | 9.66 | 香港の最高の景色を見るには、島から出て九龍のウォーターフロントに向かいましょう。 | 香港の最高の景色を見るには 島から出て九龍のウォーターフロントに向かいましょう | 香港の最高の景色を見るには、島から出て九龍のウォーターフロントに向かいましょう。 | 1.632 | 2026-07-04T13:30:06Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.976, "endtime": 7790, "rulename": "", "starttime": 1150, "tags": [], "text": "香港の最高の景色を見るには、島から出て九龍のウォーターフロントに向かいましょう。", "tokens": [{"confidence": 1.0, "endtime": 1982, "spoken": "ほんこん", "starttime": 1230, "written": "香港"}, {"confidence": 1.0, "endtime": 2094, "s... |
amivoice | -a-general | ja | ja_0010 | 11.94 | 残念ながら、運転手の行動を100%の確率で予測することはできないので、交通の流れを研究することは困難です。 | 残念ながら 運転手の行動を100%の確率で予測することはできないので 交通の流れを研究することは困難です | 残念ながら、運転手の行動を100%の確率で予測することはできないので、交通の流れを研究することは困難です。 | 2.4445 | 2026-07-04T13:30:10Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.988, "endtime": 10434, "rulename": "", "starttime": 0, "tags": [], "text": "残念ながら、運転手の行動を100%の確率で予測することはできないので、交通の流れを研究することは困難です。", "tokens": [{"confidence": 1.0, "endtime": 2232, "spoken": "ざんねん", "starttime": 1528, "written": "残念"}, {"confidence": 1.0, "endtime... |
amivoice | -a-general | ja | ja_0011 | 14.04 | 都市はモロッコのスルタンによってダルル・バディアとして再建され、貿易拠点を設立したスペイン人貿易商によってカサブランカと名付けられました。 | 都市はモロッコのスルタンによってダルル バディアとして再建され 貿易拠点を設立したスペイン人貿易商によってカサブランカと名付けられました | 年はモロッコのスルタンによってダブルバリアとして再建され、貿易拠点を設立したスペイン人貿易商によってカサブランカと名付けられました。 | 3.0668 | 2026-07-04T13:30:13Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.98, "endtime": 12240, "rulename": "", "starttime": 950, "tags": [], "text": "年はモロッコのスルタンによってダブルバリアとして再建され、貿易拠点を設立したスペイン人貿易商によってカサブランカと名付けられました。", "tokens": [{"confidence": 0.56, "endtime": 1606, "spoken": "とし", "starttime": 1302, "written": "年"}, {"confidence": 1... |
amivoice | -a-general | ja | ja_0012 | 15.06 | 復興努力に対するリベラル派からの批判では、ワシントンの内部関係者と思われる人物と復興契約を交わしたことに焦点を当てています。 | 復興努力に対するリベラル派からの批判では ワシントンの内部関係者と思われる人物と復興契約を交わしたことに焦点を当てています | 復興努力に対するリベラル派からの批判では、ワシントンの内部関係者と思われる人物と復興契約を乾かし交わしたことに焦点を当てています。 | 2.31 | 2026-07-04T13:30:16Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.999, "endtime": 13190, "rulename": "", "starttime": 950, "tags": [], "text": "復興努力に対するリベラル派からの批判では、ワシントンの内部関係者と思われる人物と復興契約を乾かし交わしたことに焦点を当てています。", "tokens": [{"confidence": 1.0, "endtime": 1846, "spoken": "ふっこう", "starttime": 1206, "written": "復興"}, {"confidence":... |
amivoice | -a-general | ja | ja_0013 | 10.08 | ロランド・メンドーサはM16ライフルで観光客に発砲しました。 | ロランド メンドーサはm16ライフルで観光客に発砲しました | ローランドメンドーサはM16ライフルで観光客に発砲しました。 | 1.4038 | 2026-07-04T13:30:18Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.993, "endtime": 6486, "rulename": "", "starttime": 1750, "tags": [], "text": "ローランドメンドーサはM16ライフルで観光客に発砲しました。", "tokens": [{"confidence": 0.71, "endtime": 2518, "spoken": "ろーらんど", "starttime": 2038, "written": "ローランド"}, {"confidence": 0.66, "endtime": 3078, "spoke... |
amivoice | -a-general | ja | ja_0014 | 11.76 | 残念ながら、運転手の行動を100%の確率で予測することはできないので、交通の流れを研究することは困難です。 | 残念ながら 運転手の行動を100%の確率で予測することはできないので 交通の流れを研究することは困難です | 残念ながら運転手の行動100%の確率で予測することはできないので交通の流れを研究することは困難です。 | 2.1083 | 2026-07-04T13:30:21Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.992, "endtime": 11490, "rulename": "", "starttime": 850, "tags": [], "text": "残念ながら運転手の行動100%の確率で予測することはできないので交通の流れを研究することは困難です。", "tokens": [{"confidence": 1.0, "endtime": 1826, "spoken": "ざんねん", "starttime": 1170, "written": "残念"}, {"confidence": 1.0, "endtime"... |
amivoice | -a-general | ja | ja_0015 | 18.3 | 私たちは植物で家を造り、植物から衣類を作ります。日々食べる食材の多くが植物です。植物がなければ動物は生きていけません。 | 私たちは植物で家を造り 植物から衣類を作ります 日々食べる食材の多くが植物です 植物がなければ動物は生きていけません | 私達は植物で家を作り、植物から衣類を作ります。日々食べる食材の多くは植物です。植物がなければ、諸動物は生きていけません。 | 3.5457 | 2026-07-04T13:30:25Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.99499995, "endtime": 15240, "rulename": "", "starttime": 0, "tags": [], "text": "私達は植物で家を作り、植物から衣類を作ります。日々食べる食材の多くは植物です。植物がなければ、諸動物は生きていけません。", "tokens": [{"confidence": 1.0, "endtime": 4790, "spoken": "わたしたち", "starttime": 3942, "written": "私達"}, {"confidence": ... |
amivoice | -a-general | ja | ja_0016 | 18.6 | 78ある提言うち最初に挙げられているのは、敵対的な介入に対してイラクの国境を確保し、近隣諸国との外交関係を再び確立するために、年内に新たな外交主導権を発揮するべきだというものです。 | 78ある提言うち最初に挙げられているのは 敵対的な介入に対してイラクの国境を確保し 近隣諸国との外交関係を再び確立するために 年内に新たな外交主導権を発揮するべきだというものです | 78ある提言うち最初に挙げられているのは、敵対的な介入に対して、イラクの国境を確保し、近隣諸国との外交関係を再び確立するために、年内に新たな外交主導権を破棄するべきだというものです。 | 3.6848 | 2026-07-04T13:30:30Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.92599994, "endtime": 16786, "rulename": "", "starttime": 0, "tags": [], "text": "78ある提言うち最初に挙げられているのは、敵対的な介入に対して、イラクの国境を確保し、近隣諸国との外交関係を再び確立するために、年内に新たな外交主導権を破棄するべきだというものです。", "tokens": [{"confidence": 1.0, "endtime": 2686, "spoken": "ななじゅうはち", "starttime": 1886, ... |
amivoice | -a-general | ja | ja_0017 | 18.3 | ドイツの焼き菓子の多くは、アーモンドやヘーゼルナッツなどの木の実を使っています。人気のあるケーキは、特に強いコーヒーとよく合います。 | ドイツの焼き菓子の多くは アーモンドやヘーゼルナッツなどの木の実を使っています 人気のあるケーキは 特に強いコーヒーとよく合います | ドイツの焼き菓子の多くは、アーモンドやヘーゼルナッツなどの木の実を使っています。人気のあるケーキは特に強いコーヒーとよく合います。 | 2.3917 | 2026-07-04T13:30:33Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.99399996, "endtime": 16886, "rulename": "", "starttime": 2950, "tags": [], "text": "ドイツの焼き菓子の多くは、アーモンドやヘーゼルナッツなどの木の実を使っています。人気のあるケーキは特に強いコーヒーとよく合います。", "tokens": [{"confidence": 1.0, "endtime": 3750, "spoken": "どいつ", "starttime": 3190, "written": "ドイツ"}, {"confid... |
amivoice | -a-general | ja | ja_0018 | 11.88 | ヤギの家畜化は、約1万年前に始まり、その中心はイランのザグロス山脈西部であったとされています。 | ヤギの家畜化は 約1万年前に始まり その中心はイランのザグロス山脈西部であったとされています | ヤギの家畜が約1年万年前に始まり、その中心は、イランのザグロス300セーブであったとされています。 | 1.9631 | 2026-07-04T13:30:36Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.9983334, "endtime": 9942, "rulename": "", "starttime": 0, "tags": [], "text": "ヤギの家畜が約1年万年前に始まり、その中心は、イランのザグロス300セーブであったとされています。", "tokens": [{"confidence": 0.7, "endtime": 2254, "spoken": "やぎ", "starttime": 1902, "written": "ヤギ"}, {"confidence": 1.0, "endtime": ... |
amivoice | -a-general | ja | ja_0019 | 12.42 | 昨日の朝、トルコのガジアンテップの警察本部で自動車爆弾の爆発により、警官2人が死亡し、負傷者は20人を超えました。 | 昨日の朝 トルコのガジアンテップの警察本部で自動車爆弾の爆発により 警官2人が死亡し 負傷者は20人を超えました | 昨日の朝、トルコのガジアンテップの警察本部で自動車爆弾の爆発により経営管理名が死亡し、負傷者は20人を超えました。 | 1.9939 | 2026-07-04T13:30:38Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.984, "endtime": 10382, "rulename": "", "starttime": 1950, "tags": [], "text": "昨日の朝、トルコのガジアンテップの警察本部で自動車爆弾の爆発により経営管理名が死亡し、負傷者は20人を超えました。", "tokens": [{"confidence": 0.97, "endtime": 2606, "spoken": "きのう", "starttime": 2238, "written": "昨日"}, {"confidence": 1.0, "... |
amivoice | -a-general | ja | ja_0020 | 11.76 | ビデオ録画の利用は、数ミリ秒程度続く微表情、つまり顔の動きを解釈するうえで重要な発見につながっています。 | ビデオ録画の利用は 数ミリ秒程度続く微表情 つまり顔の動きを解釈するうえで重要な発見につながっています | ビデオ録画の利用は、数ミリ秒程度続く日表情つまり顔の動きを解釈する上で、重要な発見に繋がっています。 | 2.0789 | 2026-07-04T13:30:41Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.9963333, "endtime": 9740, "rulename": "", "starttime": 0, "tags": [], "text": "ビデオ録画の利用は、数ミリ秒程度続く日表情つまり顔の動きを解釈する上で、重要な発見に繋がっています。", "tokens": [{"confidence": 0.99, "endtime": 2022, "spoken": "びでお", "starttime": 1414, "written": "ビデオ"}, {"confidence": 0.99, "endti... |
amivoice | -a-general | ja | ja_0021 | 15 | スンダルバンスは世界最大の沿岸マングローブ地帯であり、海岸からバングラデシュとインドの内陸地域まで80 km(50マイル)にわたって広がっています。 | スンダルバンスは世界最大の沿岸マングローブ地帯であり 海岸からバングラデシュとインドの内陸地域まで80 km50マイルにわたって広がっています | 済んだルパン諏訪世界最大の沿岸マングローブ地帯であり、海外からバングラデシュとインドの内陸地域まで80km、50マイルにわたって広がっています。 | 2.4399 | 2026-07-04T13:30:44Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.991, "endtime": 13188, "rulename": "", "starttime": 1050, "tags": [], "text": "済んだルパン諏訪世界最大の沿岸マングローブ地帯であり、海外からバングラデシュとインドの内陸地域まで80km、50マイルにわたって広がっています。", "tokens": [{"confidence": 0.58, "endtime": 1722, "spoken": "すん", "starttime": 1370, "written": "済ん"}, {"confi... |
amivoice | -a-general | ja | ja_0022 | 8.4 | これは、一部の動詞と目的語を区別するうえで重要な方法です。 | これは 一部の動詞と目的語を区別するうえで重要な方法です | これは一部の動詞の目的語を区別する上で重要な方法です。 | 1.5109 | 2026-07-04T13:30:47Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.974, "endtime": 6538, "rulename": "", "starttime": 1850, "tags": [], "text": "これは一部の動詞の目的語を区別する上で重要な方法です。", "tokens": [{"confidence": 1.0, "endtime": 2458, "spoken": "これ", "starttime": 2170, "written": "これ"}, {"confidence": 1.0, "endtime": 2618, "spoken": "は", "s... |
amivoice | -a-general | ja | ja_0023 | 8.22 | 羽を持ち、空を飛べるので、恐竜だと思っていない人も大勢います。 | 羽を持ち 空を飛べるので 恐竜だと思っていない人も大勢います | 羽を持ち空を飛べるので、恐竜だと思っていない人も当然います。 | 1.8447 | 2026-07-04T13:30:49Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.995, "endtime": 7996, "rulename": "", "starttime": 1500, "tags": [], "text": "羽を持ち空を飛べるので、恐竜だと思っていない人も当然います。", "tokens": [{"confidence": 0.82, "endtime": 2188, "spoken": "はね", "starttime": 1852, "written": "羽"}, {"confidence": 0.75, "endtime": 2300, "spoken": "を"... |
amivoice | -a-general | ja | ja_0024 | 14.46 | 同じように、シェンゲンビザさえ取得すれば、シェンゲン加盟国のビザを個別に申請する必要がなくなり、時間、費用、事務処理の手間が相当に省けます。 | 同じように シェンゲンビザさえ取得すれば シェンゲン加盟国のビザを個別に申請する必要がなくなり 時間 費用 事務処理の手間が相当に省けます | 同じように宣言ビザさえ取得すれば、信玄加盟国のビザを個別に申請する必要がなくなり、時間、費用、事務処理の手間が相当に省けます。 | 2.7462 | 2026-07-04T13:30:53Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.99549997, "endtime": 12786, "rulename": "", "starttime": 950, "tags": [], "text": "同じように宣言ビザさえ取得すれば、信玄加盟国のビザを個別に申請する必要がなくなり、時間、費用、事務処理の手間が相当に省けます。", "tokens": [{"confidence": 1.0, "endtime": 1686, "spoken": "おなじ", "starttime": 1094, "written": "同じ"}, {"confidence... |
amivoice | -a-general | ja | ja_0025 | 9.9 | 乾いた粉を2つ合わせてくるくるとこね回し、清潔な濡れた手でぎゅっと握ります。 | 乾いた粉を2つ合わせてくるくるとこね回し 清潔な濡れた手でぎゅっと握ります | かわいいと粉を二つ合わせてくるくるとこね回し、清潔な抜いた手でギュッと握ります。 | 1.9676 | 2026-07-04T13:30:55Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.976, "endtime": 7536, "rulename": "", "starttime": 1600, "tags": [], "text": "かわいいと粉を二つ合わせてくるくるとこね回し、清潔な抜いた手でギュッと握ります。", "tokens": [{"confidence": 0.86, "endtime": 2336, "spoken": "かわいい", "starttime": 1904, "written": "かわいい"}, {"confidence": 0.59, "endtime": 2432... |
amivoice | -a-general | ja | ja_0026 | 8.1 | 地殻の厚さは、手前側が約70 km、奥側が約100 kmです。 | 地殻の厚さは 手前側が約70 km 奥側が約100 kmです | 地殻の厚さは手前側が約70km、横川が約100kmです。 | 2.1232 | 2026-07-04T13:30:58Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.953, "endtime": 6540, "rulename": "", "starttime": 1500, "tags": [], "text": "地殻の厚さは手前側が約70km、横川が約100kmです。", "tokens": [{"confidence": 0.74, "endtime": 2188, "spoken": "ちかく", "starttime": 1788, "written": "地殻"}, {"confidence": 1.0, "endtime": 2284, "spoken": "の",... |
amivoice | -a-general | ja | ja_0027 | 11.76 | 生地の温度が上がりすぎないように気を付けてください(縮みの原因になったり、極端な場合は焦げ付きの原因になります)。 | 生地の温度が上がりすぎないように気を付けてください縮みの原因になったり 極端な場合は焦げ付きの原因になります | 生地の温度が上がりすぎないように気をつけてください。縮みの原因になったり、極端な場合は焦げ付けの原因になります。 | 2.0909 | 2026-07-04T13:31:01Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.995, "endtime": 9442, "rulename": "", "starttime": 1250, "tags": [], "text": "生地の温度が上がりすぎないように気をつけてください。縮みの原因になったり、極端な場合は焦げ付けの原因になります。", "tokens": [{"confidence": 0.85, "endtime": 1938, "spoken": "きじ", "starttime": 1602, "written": "生地"}, {"confidence": 1.0, "end... |
amivoice | -a-general | ja | ja_0028 | 11.64 | 結果として、舞台上で出演者が大麻を吸い、劇場側も観客に参加を呼びかけています。 | 結果として 舞台上で出演者が大麻を吸い 劇場側も観客に参加を呼びかけています | 結果として、舞台上で出演者が大麻を吸い、劇場側も観客に参加を呼びかけています。 | 2.6451 | 2026-07-04T13:31:04Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.98466665, "endtime": 10942, "rulename": "", "starttime": 400, "tags": [], "text": "結果として、舞台上で出演者が大麻を吸い、劇場側も観客に参加を呼びかけています。", "tokens": [{"confidence": 1.0, "endtime": 1152, "spoken": "けっか", "starttime": 736, "written": "結果"}, {"confidence": 1.0, "endtime": 1296, ... |
amivoice | -a-general | ja | ja_0029 | 11.88 | ダンディー大学のパメラ・ファーガソン教授は、「容疑者の写真などを公開する報道記者は危い橋を渡っている」と指摘しています。 | ダンディー大学のパメラ ファーガソン教授は 容疑者の写真などを公開する報道記者は危い橋を渡っている と指摘しています | なんで大学のパメラファーガソン教授は、容疑者の写真などを公開する報道記者は危ない橋を渡っていると指摘しています。 | 2.3373 | 2026-07-04T13:31:07Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.995, "endtime": 9646, "rulename": "", "starttime": 1550, "tags": [], "text": "なんで大学のパメラファーガソン教授は、容疑者の写真などを公開する報道記者は危ない橋を渡っていると指摘しています。", "tokens": [{"confidence": 0.61, "endtime": 2302, "spoken": "なんで", "starttime": 1902, "written": "なんで"}, {"confidence": 1.0, "e... |
amivoice | -a-general | ja | ja_0030 | 15.18 | 女子シッティンググループでは4人のスキーヤーが滑走を終えられず、ジャイアントスラロームでは全117人中45人がランクインできなかった。 | 女子シッティンググループでは4人のスキーヤーが滑走を終えられず ジャイアントスラロームでは全117人中45人がランクインできなかった | 女子シッティンググループでは4人のスキーヤーが滑走を得られるぞ。ジャイアントスラロームでは全117人中45人がランクインできなかった。 | 2.1663 | 2026-07-04T13:31:10Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.97, "endtime": 12140, "rulename": "", "starttime": 3100, "tags": [], "text": "女子シッティンググループでは4人のスキーヤーが滑走を得られるぞ。ジャイアントスラロームでは全117人中45人がランクインできなかった。", "tokens": [{"confidence": 0.58, "endtime": 3740, "spoken": "じょし", "starttime": 3420, "written": "女子"}, {"confidence... |
amivoice | -a-general | ja | ja_0031 | 10.68 | この都市は、アフリカの都市というよりもアラビア色が強く、国内の他の都市とは対照的な雰囲気です。 | この都市は アフリカの都市というよりもアラビア色が強く 国内の他の都市とは対照的な雰囲気です | この年は、アフリカの都市というよりもアラビア色が強く、国内の他の都市とは対照的な雰囲気です。 | 2.1352 | 2026-07-04T13:31:13Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 1.0, "endtime": 9388, "rulename": "", "starttime": 900, "tags": [], "text": "この年は、アフリカの都市というよりもアラビア色が強く、国内の他の都市とは対照的な雰囲気です。", "tokens": [{"confidence": 1.0, "endtime": 1556, "spoken": "この", "starttime": 1252, "written": "この"}, {"confidence": 0.91, "endtime": 1876, ... |
amivoice | -a-general | ja | ja_0032 | 18.48 | 統治者は「ビシー」のフランス人たちでした。彼らは1940年にドイツ軍と和平を結び、侵略者と戦うのではなく、侵略者に協力したフランス人たちです。 | 統治者は ビシー のフランス人たちでした 彼らは1940年にドイツ軍と和平を結び 侵略者と戦うのではなく 侵略者に協力したフランス人たちです | 当事者はわびしいのフランス人たちでした。彼らは1940年にドイツ軍と和平を結び、侵略者と戦うのではなく、侵略者に協力したフランス人たちです。 | 2.5706 | 2026-07-04T13:31:16Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.994, "endtime": 15840, "rulename": "", "starttime": 4400, "tags": [], "text": "当事者はわびしいのフランス人たちでした。彼らは1940年にドイツ軍と和平を結び、侵略者と戦うのではなく、侵略者に協力したフランス人たちです。", "tokens": [{"confidence": 0.73, "endtime": 5072, "spoken": "とうじ", "starttime": 4704, "written": "当事"}, {"confid... |
amivoice | -a-general | ja | ja_0033 | 20.76 | しかし、結核患者全体に占めるXDR-TBの割合はまだ低いようで、こうしている間にも南アフリカでは33万人のうち6,000人が感染している。 | しかし 結核患者全体に占めるxdr-tbの割合はまだ低いようで こうしている間にも南アフリカでは33万人のうち6,000人が感染している | しかし、結核患者全体に占めるX3RTの割合はまだ低いようで、こうしている間にも南アフリカでは33万人のうち6000人が感染している。 | 2.4676 | 2026-07-04T13:31:19Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.9555, "endtime": 16238, "rulename": "", "starttime": 2750, "tags": [], "text": "しかし、結核患者全体に占めるX3RTの割合はまだ低いようで、こうしている間にも南アフリカでは33万人のうち6000人が感染している。", "tokens": [{"confidence": 1.0, "endtime": 3774, "spoken": "しかし", "starttime": 3102, "written": "しかし"}, {"confidenc... |
amivoice | -a-general | ja | ja_0034 | 10.8 | 実地に体験したければ、早めに到着して演奏が聞こえやすいキャンプ場を確保しなければなりません。 | 実地に体験したければ 早めに到着して演奏が聞こえやすいキャンプ場を確保しなければなりません | 実地に体験したければ早めに到着して演奏が聞こえやすいキャンプ場を確保しなければなりません。 | 2.2256 | 2026-07-04T13:31:22Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.985, "endtime": 10482, "rulename": "", "starttime": 1250, "tags": [], "text": "実地に体験したければ早めに到着して演奏が聞こえやすいキャンプ場を確保しなければなりません。", "tokens": [{"confidence": 1.0, "endtime": 2066, "spoken": "じっち", "starttime": 1554, "written": "実地"}, {"confidence": 0.52, "endtime": 22... |
amivoice | -a-general | ja | ja_0035 | 16.44 | 香港の地平線を描いて立ち並ぶビル群は、ビクトリア・ハーバーの水の存在によって、きらびやかな棒グラフに例えられています。 | 香港の地平線を描いて立ち並ぶビル群は ビクトリア ハーバーの水の存在によって きらびやかな棒グラフに例えられています | 香港の地平線を描いて立ち並ぶビル群はビクトリアハーバーの水の存在によって、きらびやかな棒グラフに例えられています。 | 3.1528 | 2026-07-04T13:31:26Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.9943333, "endtime": 15792, "rulename": "", "starttime": 1050, "tags": [], "text": "香港の地平線を描いて立ち並ぶビル群はビクトリアハーバーの水の存在によって、きらびやかな棒グラフに例えられています。", "tokens": [{"confidence": 1.0, "endtime": 1978, "spoken": "ほんこん", "starttime": 1082, "written": "香港"}, {"confidence": 1.... |
amivoice | -a-general | ja | ja_0036 | 9.3 | 洞窟はメッカの北にあたる山頂にあり、外界から完全に隔離されています。 | 洞窟はメッカの北にあたる山頂にあり 外界から完全に隔離されています | 洞窟はメッカの北にあたる山頂にあり、下界から完全に隔離されています。 | 1.6528 | 2026-07-04T13:31:29Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.99950004, "endtime": 7790, "rulename": "", "starttime": 1550, "tags": [], "text": "洞窟はメッカの北にあたる山頂にあり、下界から完全に隔離されています。", "tokens": [{"confidence": 1.0, "endtime": 2382, "spoken": "どうくつ", "starttime": 1822, "written": "洞窟"}, {"confidence": 0.93, "endtime": 2478, "s... |
amivoice | -a-general | ja | ja_0037 | 15.6 | プライドは、1~3頭の血縁関係にあるオスと30頭余りのメスと子供たちで構成されます。 | プライドは 1~3頭の血縁関係にあるオスと30頭余りのメスと子供たちで構成されます | プライドは1から3頭の血縁関係にある雄と30頭余りのメスと子供たちで構成されます。 | 1.8083 | 2026-07-04T13:31:31Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.997, "endtime": 11934, "rulename": "", "starttime": 3950, "tags": [], "text": "プライドは1から3頭の血縁関係にある雄と30頭余りのメスと子供たちで構成されます。", "tokens": [{"confidence": 0.87, "endtime": 4846, "spoken": "ぷらいど", "starttime": 4238, "written": "プライド"}, {"confidence": 0.99, "endtime": 51... |
amivoice | -a-general | ja | ja_0038 | 17.46 | 彼は多くの国で紙幣の製作にも携わっており、最近では、カナダの新5ドル紙幣と新100ドル紙幣の表に描かれた首相の肖像画などがその一例です。 | 彼は多くの国で紙幣の製作にも携わっており 最近では カナダの新5ドル紙幣と新100ドル紙幣の表に描かれた首相の肖像画などがその一例です | 彼は多くの国で紙幣の製作にも携わっており、最近ではカナダの新5ドル紙幣と新100ドル紙幣の表に書かれた。輸送の焦燥がなどがその一例です。 | 3.3031 | 2026-07-04T13:31:35Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.954, "endtime": 14340, "rulename": "", "starttime": 3600, "tags": [], "text": "彼は多くの国で紙幣の製作にも携わっており、最近ではカナダの新5ドル紙幣と新100ドル紙幣の表に書かれた。輸送の焦燥がなどがその一例です。", "tokens": [{"confidence": 1.0, "endtime": 4240, "spoken": "かれ", "starttime": 3968, "written": "彼"}, {"confidence"... |
amivoice | -a-general | ja | ja_0039 | 17.94 | その長い顎には70本以上の鋭い歯が並び、口蓋には別の歯列があり、つまりここを通ったら逃げ道はないということになります。 | その長い顎には70本以上の鋭い歯が並び 口蓋には別の歯列があり つまりここを通ったら逃げ道はないということになります | その長い顎には70本以上の鋭い歯が並び、郊外には別の歯列があり、つまりここを通ったら逃げ道はないということになります。 | 2.4564 | 2026-07-04T13:31:38Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.99399996, "endtime": 14446, "rulename": "", "starttime": 3400, "tags": [], "text": "その長い顎には70本以上の鋭い歯が並び、郊外には別の歯列があり、つまりここを通ったら逃げ道はないということになります。", "tokens": [{"confidence": 1.0, "endtime": 4104, "spoken": "その", "starttime": 3704, "written": "その"}, {"confidence": 1... |
amivoice | -a-general | ja | ja_0040 | 17.82 | 一般的に言えば、管理職がかつての仲間を率いることになった場合、2とおりの行動が現れる可能性があります。1つは「仲間の1人」のままでいようとする行動です。 | 一般的に言えば 管理職がかつての仲間を率いることになった場合 2とおりの行動が現れる可能性があります 1つは 仲間の1人 のままでいようとする行動です | 一般的に言えば、管理職はかつての仲間を率いることになった場合、2通りの行動が現れる可能性があります。一つは、仲間の1人のままでいようとする行動です。 | 3.3023 | 2026-07-04T13:31:42Z | 0 | null | {"audio_conversion": "float32 wav -> pcm_16 wav (same samples)", "endpoint": "https://acp-api.amivoice.com/v1/recognize", "grammarFileNames": "-a-general", "logging": true} | {"code": "", "message": "", "results": [{"confidence": 0.94, "endtime": 15246, "rulename": "", "starttime": 3650, "tags": [], "text": "一般的に言えば、管理職はかつての仲間を率いることになった場合、2通りの行動が現れる可能性があります。一つは、仲間の1人のままでいようとする行動です。", "tokens": [{"confidence": 1.0, "endtime": 4466, "spoken": "いっぱん", "starttime": 3986, "written": "一般"}, {"con... |
koedesk STT Bench — raw transcriptions (v1, Track A)
28,400 raw transcriptions: 13 hosted speech-to-text engine variants × 12 languages × 200 FLEURS utterances, produced by a preregistered benchmark run on 2026-07-05. Every engine output is published unmodified, including full raw API responses.
Conflict of interest, stated first: this benchmark is built and funded by koedesk, a voice-typing product that uses ElevenLabs Scribe v2. We do not pretend to be neutral. Instead, everything you need to distrust us and re-score the data yourself is public: the methodology was frozen in PREREGISTRATION.md before the headline run (verifiable from git history), all raw outputs are in this dataset, and the scoring code is MIT-licensed on GitHub. Where Scribe loses, the result is published unchanged — it is not #1 overall.
- Leaderboard: https://koedesk.app/benchmark/
- Methodology, scoring code, normalizers, amendments: https://github.com/guide-inc-org/koedesk-stt-bench
- Write-ups: English · 日本語 (Zenn)
Configs
transcriptions (28,400 rows)
One row per (engine variant, language, utterance).
| column | type | description |
|---|---|---|
engine |
string | Engine variant key (matches scores.json / leaderboard) |
model_id |
string | Exact model identifier sent to the API |
lang |
string | Benchmark language code (see FLEURS mapping below) |
utt_id |
string | {lang}_{index:04d} — index into the FLEURS test split, dataset order |
duration_sec |
float | Audio duration |
ref_text_raw |
string | FLEURS reference transcription, untouched |
ref_text_normalized |
string | Reference after the preregistered normalization pipeline |
hypothesis |
string | Engine output text, untouched |
latency_sec |
float | Wall-clock API latency for this utterance |
timestamp_utc |
string | When the API call was made |
retries |
int | Retry count for this utterance |
error |
string/null | Final error, if any (all rows in v1 completed successfully) |
params_json |
string | JSON: exact request parameters |
raw_response_json |
string | JSON: the full, untruncated API response |
cell_scores (142 rows)
One row per (engine, language) cell, computed by the frozen scoring code (make_scores.py); scores.json is also included verbatim at the repo root. Columns include corpus CER/WER, utterance means, bootstrap 95% CIs, real-time-factor percentiles, list price per audio hour, and the tie group (tie_group, 1 = best tier). The headline metric is CER for ja/th/zh and WER elsewhere (headline_metric column); wer_* is null for th (no whitespace word boundaries).
Read results through tie groups, not raw ranks. Within a tie group, pairwise differences are not statistically distinguishable (paired bootstrap, preregistered in §6). Claiming "engine A beats engine B" inside the same tier is exactly the misuse this benchmark tries to prevent.
Engines (13 hosted variants)
| engine | model_id | languages |
|---|---|---|
elevenlabs |
scribe_v2 | 12 |
openai |
gpt-4o-transcribe | 12 |
whisper1 |
whisper-1 | 12 |
google_stt_v2 |
chirp_3 | 12 |
deepgram |
nova-3 | 12 |
mai_transcribe |
mai-transcribe-1.5 | 12 |
gemini |
gemini-2.5-flash | 12 |
gemini_25_pro |
gemini-2.5-pro | 12 |
gemini_31_flash_lite |
gemini-3.1-flash-lite | 12 |
gemini_31_pro |
gemini-3.1-pro-preview | 12 |
gemini_35_flash |
gemini-flash-latest | 12 |
mistral_voxtral |
voxtral-mini-2602 | 9 (no id/th/vi) |
amivoice |
-a-general | 1 (ja only) |
Engine selection and exclusions (e.g. engines whose terms of service prohibit publishing benchmark results) are documented in the ToS audit.
Audio provenance (audio is NOT redistributed here)
Test material is the google/fleurs test split, first 200 utterances in dataset order per language (deterministic — no sampling choices). To reconstruct the exact audio: load the config below, take rows 0–199 of the test split, then apply normalize_audio.py (16-bit PCM, −6 dBFS peak — the identical file was sent to every engine).
| lang | FLEURS config | lang | FLEURS config |
|---|---|---|---|
| ja | ja_jp | ko | ko_kr |
| en | en_us | vi | vi_vn |
| zh | cmn_hans_cn | id | id_id |
| de | de_de | th | th_th |
| fr | fr_fr | es | es_419 |
| pt | pt_br | ru | ru_ru |
Known artifacts (kept as measured, per the preregistration)
gemini_25_pro/zh/zh_0192: the model leaked ~4,800 characters of English reasoning into the transcription field for one utterance. That single utterance accounts for 86.7% of the cell's error mass (CER 48.3% as measured; 6.45% excluding it). This is the engine's real behavior, not a pipeline bug, and is published as measured.- FLEURS zh references: 13/200 reference texts contain English annotation notes from the source dataset; this affects all engines identically.
- Gemini-family latency: measured through Cloudflare AI Gateway (affects
latency_sec/RTF for the fivegemini*variants only; accuracy is unaffected). All other engines were called directly. - Measurement environment: run from Vietnam with an AWS Tokyo exit node (see Amendment 6).
Limitations
This is Track A only: FLEURS is public and every commercial engine has plausibly trained on it. These numbers connect to vendor-claimed figures and the HF Open ASR Leaderboard; they do not prove real-world dictation accuracy. A planned Track B (fresh recorded audio) was cancelled before any measurement existed — see Amendment 7. The benchmark refreshes quarterly (§9 of the preregistration).
License and attribution
- This dataset (transcriptions, scores, documentation): CC BY 4.0, © 2026 Guide Inc. (koedesk).
- Reference texts (
ref_text_raw,ref_text_normalized) derive from FLEURS (google/fleurs, CC BY 4.0; Conneau et al., 2022, FLEURS: Few-shot Learning Evaluation of Universal Representations of Speech). - Engine outputs are published for evaluation/research purposes; each provider's terms were audited before inclusion (see ToS audit).
Citation
@misc{koedesk-stt-bench-2026,
title = {koedesk STT Bench: a preregistered multilingual speech-to-text benchmark},
author = {{Guide Inc. (koedesk)}},
year = {2026},
url = {https://koedesk.app/benchmark/},
note = {Raw data: https://huggingface.co/datasets/koedesk/stt-bench. Methodology: https://github.com/guide-inc-org/koedesk-stt-bench}
}
- Downloads last month
- 40