Update app.py with fixed imports

app.py

@@ -27,7 +27,7 @@ else:
     sst_model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-base-960h")
     sst_processor = Wav2Vec2Processor.from_pretrained("facebook/wav2vec2-base-960h")
 
-# LLM: Use local GGUF file if available, else raise error (must be uploaded)
+# LLM: Use local GGUF file if available, else raise error
 if os.path.exists("./models/llama.gguf"):
     llm = Llama("./models/llama.gguf")
 else:
@@ -43,7 +43,6 @@ class LLMRequest(BaseModel):
 # API Endpoints
 @app.post("/tts")
 async def tts_endpoint(request: TTSRequest):
-    """Convert text to speech and return audio."""
     text = request.text
     inputs = tts_tokenizer(text, return_tensors="pt")
     with torch.no_grad():
@@ -56,7 +55,6 @@ async def tts_endpoint(request: TTSRequest):
 
 @app.post("/sst")
 async def sst_endpoint(file: UploadFile = File(...)):
-    """Convert speech to text and return transcription."""
     audio_bytes = await file.read()
     audio, sr = sf.read(io.BytesIO(audio_bytes))
     inputs = sst_processor(audio, sampling_rate=sr, return_tensors="pt")
@@ -68,7 +66,6 @@ async def sst_endpoint(file: UploadFile = File(...)):
 
 @app.post("/llm")
 async def llm_endpoint(request: LLMRequest):
-    """Generate text from a prompt using Llama.cpp."""
     prompt = request.prompt
     output = llm(prompt, max_tokens=50)
     return {"text": output["choices"][0]["text"]}