Azure Whisper无法识别语音
Closed this issue · 6 comments
docker-compose配置：
Options: [openai-whisper, azure-whisper, azure-stt]
STT_TYPE: azure-whisper
Azure Whisper
AZURE_WHISPER_API_BASE: https://northcentralus.api.cognitive.microsoft.com/
AZURE_WHISPER_KEY: **************
AZURE_WHISPER_DEPLOYMENT_NAME: whisper
AZURE_WHISPER_API_VERSION: 2023-09-01-preview
日志显示：
Starting new HTTPS connection (1): northcentralus.api.cognitive.microsoft.com:443
folotoy-folotoy-1 | 2024-01-05 11:08:51,250 - DEBUG - https://northcentralus.api.cognitive.microsoft.com:443 "POST //openai/deployments/whisper/audio/transcriptions?api-version=2023-09-01-preview HTTP/1.1" 404 198
folotoy-folotoy-1 | 2024-01-05 11:08:51,252 - DEBUG - [Dkey=F234103024] STT(azure-whisper) request time cost: 1.74s
folotoy-folotoy-1 | 2024-01-05 11:08:51,252 - ERROR - LLM error: Traceback (most recent call last):
folotoy-folotoy-1 | File "core/speech_wav_processor.py", line 100, in core.speech_wav_processor.SpeechWavProcessor.write_wav
folotoy-folotoy-1 | KeyError: 'text'
folotoy-folotoy-1 | Traceback (most recent call last):
folotoy-folotoy-1 | File "core/speech_wav_processor.py", line 100, in core.speech_wav_processor.SpeechWavProcessor.write_wav
folotoy-folotoy-1 | KeyError: 'text'
我反复确认，key是没错的
AZURE_WHISPER_API_BASE 不是用 cognitive 的地址，用 Azure OpenAI 的 url，例如部署的 Azure OpenAI 资源名字是 xxxx，那么 url 是：
https://xxxx.openai.azure.com/
可是我的地址并不是openai.azure.com啊
这个endpoint是分配的，无法修改
需要正确填写自己部署的 model 的 deployment。