Update apis/chat_api.py

apis/chat_api.py

@@ -313,7 +313,51 @@ async def whisper_transcribe(
     language: str = Form(),
     model: str = Form(),
 ):
-    return {"filename": audio_file.content_type}
+    MODEL_MAP = {
+        "whisper-small": "openai/whisper-small",
+        "whisper-medium": "openai/whisper-medium",
+        "whisper-large": "openai/whisper-large",   
+        "default": "openai/whisper-small",
+    }
+    AUDIO_MAP = {
+        "audio/wav": "audio/wav",
+        "audio/mpeg": "audio/mpeg",
+        "audio/x-flac": "audio/x-flac",   
+    }
+    item_response = {
+            "statue": 200,
+            "result": "",
+            "start": 0,
+            "end": 0
+    }
+    if audio_file.content_type in AUDIO_MAP.keys():
+        if model in MODEL_MAP.keys():
+            target_model = model
+        else:
+            target_model = "default"
+
+        real_name = MODEL_MAP[target_model]
+        device = 0 if torch.cuda.is_available() else "cpu"
+        pipe = pipeline(
+           task="automatic-speech-recognition",
+           model=real_name,
+           chunk_length_s=30,
+           device=device,
+        )
+        time_start = time.time()
+        pipe.model.config.forced_decoder_ids = pipe.tokenizer.get_decoder_prompt_ids(language=language, task="transcribe")
+        text = pipe(audio_file)["text"]
+        time_end = time.time()
+        item_response["status"] = 200
+        item_response["result"] = text
+        item_response["start"] = time_start
+        item_response["end"] = time_end
+    else:
+        item_response["status"] = 400
+        item_response["result"] = 'Acceptable files: audio/wav,audio/mpeg,audio/x-flac'
+        
+    
+    return item_response
     
 if __name__ == "__main__":
     args = ArgParser().args