Env-TTS-Clean

Environment-aware text-to-speech training corpus (clean release). Each row pairs four short 24 kHz mono FLAC clips with aligned transcripts:

• an environment sample (different speaker, same acoustic scene),
• a speaker reference (same speaker as the target utterance),
• a speaker-enhanced copy of the reference (MossFormer2 speech enhancement),
• the target speech to synthesise,

so a model can learn to generate an utterance with both a specified voice and a specified environment. This release supersedes the earlier Env-TTS-SD-Corpus with a richer schema, higher sample rate, per-clip ASR for all three contexts, and two additional meeting sources (AMI, AliMeeting).

Dataset statistics

metric	value
rows	189,918
on-disk size	~119 GB
Σ speech_duration	1,542,868.6 s (428.6 h)
Σ (environment_audio_duration + speaker_audio_duration + speech_duration)	4,507,421.5 s (1,252.1 h)

Duration totals were computed on 2026-05-27 by summing the float duration columns across all 239 parquet shards (no audio decode).

Rows by source dataset

dataset	rows
m3sd	108,708
aishell4	27,924
alimeeting	25,087
ami	13,207
msdwild	9,676
chime6	5,316

Schema

column	type	description
environment_audio_source	binary (FLAC 24 kHz mono)	acoustic-scene reference, 2.5–15 s, from a different speaker in the same session
environment_audio_duration	float32	seconds
environment_audio_text	string	transcript of the environment clip (gold or Qwen3-ASR)
speaker_audio_source	binary (FLAC 24 kHz mono)	speaker-identity reference, 2.5–15 s, same speaker as speech
speaker_audio_duration	float32	seconds
speaker_audio_text	string	transcript of the speaker reference clip
speaker_audio_source_enhanced	binary (FLAC 24 kHz mono)	MossFormer2-enhanced version of speaker_audio_source
text	string	transcript of speech
speech	binary (FLAC 24 kHz mono)	target utterance, 3–15 s
speech_duration	float32	seconds
language	string	zh / en / auto
dataset	string	m3sd / aishell4 / msdwild / chime6 / ami / alimeeting
conversation_id	string	unique within the source dataset
speaker_id	string	within-conversation diarisation label
env_id	string	acoustic-scene identifier (usually conversation_id)
text_source	string	original, asr, or mixed
asr_token_count	int32	Qwen3-ASR token count for speech
asr_mean_logprob	float32	mean log-prob per token for speech

Source corpora

dataset	nominal hours	language	transcripts in corpus
M3SD (Wu et al., 2025)	770	zh / en mixed	❌ → Qwen3-ASR
AISHELL-4 (Fu et al., 2021)	120	zh	✅ TextGrid
MSDWILD (Liu et al., 2022)	80	zh / en mixed	❌ → Qwen3-ASR
CHiME-6 (Watanabe et al., 2020)	40+	en	✅ JSON
AMI (SDM, diarizers-community)	~100	en	❌ → Qwen3-ASR
AliMeeting (OpenSLR 119, far ch.0)	~120	zh	✅ TextGrid

Processing pipeline

Built with the streaming pipeline in env-tts-data-pipeline (three parallel stages: download → process → upload):

1. download — stream each source conversation (HF mirrors, OpenSLR tar streams, BaiduPCS for M3SD, etc.) into a bounded local cache; emit a JSON sentinel when audio is ready.
2. process — resample to 24 kHz mono, walk diarisation turns, emit 3–15 s speech slices with a same-speaker reference (≥2.5 s) and a different-speaker environment slice (≥2.5 s, extended into surrounding audio when needed). Missing or split transcripts are re-labelled with Qwen3-ASR-1.7B (vLLM backend). Rows are written as snappy parquet shards (~800 rows / shard, 4 shards per HF commit group).
3. upload — HfApi.upload_folder per sealed group with rate limiting and resume-safe JSON state under state/.
4. enhance (second pass) — stream each published shard back, run MossFormer2_SE_48K on speaker_audio_source, and fill speaker_audio_source_enhanced at the same 24 kHz storage rate.

DAIEN-style RIR+noise augmentation on the speaker reference is disabled in this release; environment/speaker decoupling relies on cross-speaker env sampling plus MossFormer2 enhancement instead.

Licensing

Released under CC-BY-SA-4.0, inheriting the most restrictive terms among sources. In particular:

• M3SD — academic / non-commercial research only.
• MSDWILD — X-LANCE research-only agreement.
• AISHELL-4 (Apache-2.0), CHiME-6 (CC-BY-SA-4.0), AMI, and AliMeeting carry their respective open/research terms.

Redistributing extracted audio requires complying with each upstream licence.

Citation

Please cite the source papers when using this corpus:

@article{wu2025m3sd,
  title={M3SD: Multi-modal, Multi-scenario and Multi-language Speaker
         Diarization Dataset},
  author={Wu, Shilong and others},
  journal={arXiv preprint arXiv:2506.14427},
  year={2025}
}

@inproceedings{fu2021aishell4,
  title={AISHELL-4: An Open Source Dataset for Speech Enhancement, Separation,
         Recognition and Speaker Diarization in Conference Scenario},
  author={Fu, Yihui and others},
  booktitle={Interspeech},
  year={2021}
}

@inproceedings{liu2022msdwild,
  title={MSDWILD: Multi-modal Speaker Diarization Dataset in the Wild},
  author={Liu, Tao and others},
  booktitle={Interspeech},
  year={2022}
}

@inproceedings{watanabe2020chime6,
  title={CHiME-6 Challenge: Tackling Multispeaker Speech Recognition for
         Unsegmented Recordings},
  author={Watanabe, Shinji and others},
  booktitle={CHiME Workshop},
  year={2020}
}

ASR re-labelling uses Qwen3-ASR-1.7B. Speaker enhancement uses MossFormer2 (ClearVoice).

Loading

from datasets import load_dataset

ds = load_dataset("ChristianYang/Env-TTS-Clean", split="train", streaming=True)
row = next(iter(ds))
print(row["text"])
print(row["speech_duration"])
# Audio columns decode automatically when accessed (24 kHz mono).

Files on disk

data/
  group_00000/
    manifest.json
    data_000000.parquet
    data_000001.parquet
    ...
  group_00001/
    ...

Each group_* directory is one atomic HF commit bundle (typically 4 × 800-row parquet shards, snappy-compressed FLAC payloads inside).