feat: whisper使用开源模型

main.py

@@ -7,6 +7,8 @@ from langchain import OpenAI
 from langchain.chains import ConversationChain
 from langchain.memory import ConversationSummaryBufferMemory
 from langchain.chat_models import ChatOpenAI
+from transformers import pipeline
+
 
 openai.api_key = os.environ["OPENAI_API_KEY"]
 did_api_key = os.environ.get('DID_API_KEY')
@@ -100,13 +102,22 @@ def predict(input, history=[]):
         return responses, video_html, history
 
 
+# 通过openai whisper 语音识别
+# def transcribe(audio):
+#     os.rename(audio, audio + '.wav')
+#     audio_file = open(audio + '.wav', "rb")
+#     transcript = openai.Audio.transcribe(
+#         "whisper-1", audio_file, prompt="这是一段简体中文的问题。")
+#     return transcript['text']
+
+
 # 通过openai whisper 语音识别
 def transcribe(audio):
     os.rename(audio, audio + '.wav')
     audio_file = open(audio + '.wav', "rb")
-    transcript = openai.Audio.transcribe(
-        "whisper-1", audio_file, prompt="这是一段简体中文的问题。")
-    return transcript['text']
+    transcriber = pipeline(model="openai/whisper-medium", device=0)
+    result = transcriber(audio_file)
+    return result['text']
 
 
 def process_audio(audio, history=[]):

requirements.txt

@@ -1,45 +1,8 @@
-# attr==0.3.2
-# blobfile==2.0.2
-# brotli==1.0.9
-# brotlicffi==1.0.9.2
-# ConfigParser==6.0.0
-# contextlib2==21.6.0
-# cryptography==41.0.3
-# Cython==3.0.0
-# dl==0.1.0
-# docutils==0.20.1
-# gradio==3.39.0
-# HTMLParser==0.0.2
-# importlib_metadata==6.8.0
-# ipython==8.14.0
-# ipywidgets==8.1.0
-# Jinja2==3.1.2
-# jnius==1.1.0
-# keyring==24.2.0
-# langchain==0.0.249
-# lockfile==0.12.2
-# mock==5.1.0
-# numpy==1.25.2
-# openai==0.27.8
-# ordereddict==1.1
-# Pillow==10.0.0
-# protobuf==4.23.4
-# pyOpenSSL==23.2.0
-# railroad==0.5.0
-# simplejson==3.19.1
-# Sphinx==7.1.2
-# tornado==6.3.2
-# truststore==0.7.0
-# urllib3_secure_extra==0.1.0
-# # wincertstore==0.2.1
-# # xmlrpclib==1.0.1
-# zstandard==0.21.0
-# tiktoken==0.4.0
-
 openai
 langchain
 requests
 gradio
 tiktoken
+transformers