Switch to facebook/tts_transformer-en-ljspeech for TTS

app.py

@@ -1,5 +1,4 @@
-from fastapi import FastAPI, File, UploadFile, Response
-from transformers import ParlerTTSForConditionalGeneration, AutoTokenizer
+from transformers import AutoModelForSpeechSeq2Seq, AutoTokenizer
 from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
 from llama_cpp import Llama
 import torch
@@ -7,31 +6,32 @@ import soundfile as sf
 import io
 import os
 from pydantic import BaseModel
-
+from fastapi import FastAPI, File, UploadFile, Response
 app = FastAPI()
 
-# Load models
+# Load TTS model
 if os.path.exists("./models/tts_model"):
-    tts_model = ParlerTTSForConditionalGeneration.from_pretrained("./models/tts_model")
+    tts_model = AutoModelForSpeechSeq2Seq.from_pretrained("./models/tts_model")
     tts_tokenizer = AutoTokenizer.from_pretrained("./models/tts_model")
 else:
-    tts_model = ParlerTTSForConditionalGeneration.from_pretrained("parler-tts/parler-tts-mini-v1")
-    tts_tokenizer = AutoTokenizer.from_pretrained("parler-tts/parler-tts-mini-v1")
+    tts_model = AutoModelForSpeechSeq2Seq.from_pretrained("facebook/tts_transformer-en-ljspeech")
+    tts_tokenizer = AutoTokenizer.from_pretrained("facebook/tts_transformer-en-ljspeech")
 
-# SST and LLM loading remains unchanged
+# Load SST model
 if os.path.exists("./models/sst_model"):
     sst_model = Wav2Vec2ForCTC.from_pretrained("./models/sst_model")
     sst_processor = Wav2Vec2Processor.from_pretrained("./models/sst_model")
 else:
     sst_model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-base-960h")
-    sst_processor = Wav2Vec2Vec2Processor.from_pretrained("facebook/wav2vec2-base-960h")
+    sst_processor = Wav2Vec2Processor.from_pretrained("facebook/wav2vec2-base-960h")
 
+# Load LLM model
 if os.path.exists("./models/llama.gguf"):
     llm = Llama("./models/llama.gguf")
 else:
     raise FileNotFoundError("Please upload llama.gguf to models/ directory")
 
-# Request models and endpoints remain unchanged
+# Request models (unchanged)
 class TTSRequest(BaseModel):
     text: str
 
@@ -50,6 +50,7 @@ async def tts_endpoint(request: TTSRequest):
     buffer.seek(0)
     return Response(content=buffer.getvalue(), media_type="audio/wav")
 
+# SST and LLM endpoints remain unchanged
 @app.post("/sst")
 async def sst_endpoint(file: UploadFile = File(...)):
     audio_bytes = await file.read()