Used gTTS for TTS as previous TTS model didn't work

app.py

@@ -5,9 +5,11 @@ import gradio as gr
 import requests
 from datetime import datetime
 import tempfile
+from gtts import gTTS
+import os
 
 # Supabase configuration
-SUPABASE_URL = "https://gptmdbhzblfybdnohqnh.supabase.co" 
+SUPABASE_URL = "https://gptmdbhzblfybdnohqnh.supabase.co"
 SUPABASE_API_KEY = "eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJpc3MiOiJzdXBhYmFzZSIsInJlZiI6ImdwdG1kYmh6YmxmeWJkbm9ocW5oIiwicm9sZSI6ImFub24iLCJpYXQiOjE3NDc0NjY1NDgsImV4cCI6MjA2MzA0MjU0OH0.CfWArts6Kd_x7Wj0a_nAyGJfrFt8F7Wdy_MdYDj9e7U"
 SUPABASE_TABLE = "translations"
 
@@ -21,9 +23,9 @@ model_indic_to_en = AutoModelForSeq2SeqLM.from_pretrained("ai4bharat/indictrans2
 tokenizer_indic_to_en = AutoTokenizer.from_pretrained("ai4bharat/indictrans2-indic-en-1B", trust_remote_code=True)
 ip = IndicProcessor(inference=True)
 
-# Whisper STT and English TTS pipelines
+# Whisper STT pipeline (keep as is)
 asr = pipeline("automatic-speech-recognition", model="openai/whisper-small")
-tts_en = pipeline("text-to-speech", model="espnet/kan-bayashi_ljspeech_vits")
+
 
 # Save to Supabase
 def save_to_supabase(input_text, output_text, direction):
@@ -126,12 +128,15 @@ def transcribe_audio(audio_path):
         print("STT Error:", e)
         return "⚠️ Transcription failed."
 
-# Synthesize English audio if direction is ks_to_en
+# Synthesize English TTS using gTTS for ks_to_en direction
 def synthesize_tts(text, direction):
-    if direction == "ks_to_en":
+    if direction == "ks_to_en" and text.strip():
         try:
-            result = tts_en(text)
-            return (result["sampling_rate"], result["audio"])
+            tts = gTTS(text=text, lang="en")
+            tmp_file = tempfile.NamedTemporaryFile(delete=False, suffix=".mp3")
+            tts.save(tmp_file.name)
+            tmp_file.close()
+            return tmp_file.name
         except Exception as e:
             print("TTS Error:", e)
     return None

postBuild

@@ -3,5 +3,4 @@ AutoModelForSeq2SeqLM.from_pretrained('ai4bharat/indictrans2-en-indic-1B'); \
 AutoTokenizer.from_pretrained('ai4bharat/indictrans2-en-indic-1B'); \
 AutoModelForSeq2SeqLM.from_pretrained('ai4bharat/indictrans2-indic-en-1B'); \
 AutoTokenizer.from_pretrained('ai4bharat/indictrans2-indic-en-1B'); \
-pipeline('automatic-speech-recognition', model='openai/whisper-small'); \
-pipeline('text-to-speech', model='espnet/kan-bayashi_ljspeech_vits')"
+pipeline('automatic-speech-recognition', model='openai/whisper-small')"

requirements.txt

@@ -1,7 +1,7 @@
 torch>=1.12
 transformers>=4.30.0
-sentencepiece             # Required for tokenizer in IndicTrans2
-torchaudio                # Required by Whisper and ESPnet TTS
+sentencepiece
 gradio
 requests
 git+https://github.com/VarunGumma/IndicTransToolkit.git
+gTTS