whisper-small-dialect-ko
openai/whisper-smallμ νκ΅μ΄ 4κ° μ§μ λ°©μΈ(κ°μλ, κ²½μλ, μ λΌλ, μΆ©μ²λ) μμ± β νμ€μ΄ ν
μ€νΈ λ³ν νμ€ν¬λ‘ νμΈνλν λͺ¨λΈμ
λλ€.
μ¨λλ°μ΄μ€ IoT κΈ°κΈ°(νκ²: μ€λ μ§νμ΄ 5, RK3588)μμ "μ¬ν¬λ¦¬ μμ± λͺ λ Ή β νμ€μ΄ λ³ν" κΈ°λ₯μ μν΄ νμ΅λμμ΅λλ€.
νμ΅ λ°μ΄ν°
AIHub νκ΅μ΄ λ°©μΈ λ°ν λ°μ΄ν°μ (κ°μλ/κ²½μλ/μ λΌλ/μΆ©μ²λ)μμ μ§μλ³λ‘ μλμ κ°μ΄ μμ§ν λ°νλ₯Ό μ¬μ©νμ΅λλ€ (μ λΌλλ λΌλ²¨ λ°μ΄ν°κ° ꡬ쑰μ μΌλ‘ μ μ΄ μνμ μ λλ¬ν¨).
| μ§μ | νμ΅ λ°ν μ |
|---|---|
| κ²½μλ | 8,542 |
| μΆ©μ²λ | 6,310 |
| κ°μλ | 4,290 |
| μ λΌλ | 1,169 |
| ν©κ³ | 20,151 |
μ±λ₯ (CER, base whisper-small λλΉ)
| μ§μ | base | νμΈνλ |
|---|---|---|
| κ°μλ | 0.140 | 0.058 |
| κ²½μλ | 0.183 | 0.070 |
| μ λΌλ | 0.233 | 0.136 |
| μΆ©μ²λ | 0.331 | 0.121 |
| μ 체 νκ· | 0.222 | 0.096 |
μλ €μ§ νκ³
- CER κ°μ μ μ£Όλ‘ μμ μΈμ μ νλ ν₯μμμ μ¨ κ²μΌλ‘ 보μ΄λ©°, "λ¨λβλ¨λ¨ν", "μΈλ₯", "κ·Έλ°λ€λ" κ°μ λ°©μΈ νΉμ μ΄νΒ·μ΄λ―Έμ νμ€μ΄ μλ―Έ μΉνμ λ³λμ μ½μ μΌλ‘ λ¨μ μμ΅λλ€.
- μ μ£Όλ λ°©μΈμ base CERμ΄ 0.8 μμ€μΌλ‘ λ€λ₯Έ μ§μκ³Ό μ΄μ§μ μ΄μ΄μ μ΄ λͺ¨λΈμ νμ΅ λ²μμμ μ μΈλμμ΅λλ€.
μ¬μ© μμ
from transformers import WhisperForConditionalGeneration, WhisperProcessor
model = WhisperForConditionalGeneration.from_pretrained("kimyh716/whisper-small-dialect-ko")
processor = WhisperProcessor.from_pretrained("kimyh716/whisper-small-dialect-ko")
input_features = processor(audio, sampling_rate=16000, return_tensors="pt").input_features
generated_ids = model.generate(input_features, language="korean", task="transcribe")
text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
- Downloads last month
- 29
Model tree for kimyh716/whisper-small-dialect-ko
Base model
openai/whisper-small