Update app.py

app.py

@@ -12,7 +12,7 @@ tokenizer = AutoTokenizer.from_pretrained("alirezamsh/small100")
 model = AutoModelForSeq2SeqLM.from_pretrained("alirezamsh/small100")
 
 def translate_speech(audio, target_lang):
-    audio = audio.astype("float32")
+    audio = audio[0].astype("float32")  # Extract audio from tuple and convert to float32
     audio = whisper.pad_or_trim(audio, whisper_model.audio_config.sample_rate)
     mel = whisper.log_mel_spectrogram(audio).to(whisper_model.device)
     _, probs = whisper_model.detect_language(mel)
@@ -21,7 +21,7 @@ def translate_speech(audio, target_lang):
     text = result.text
 
     # Translate text
-    tokenizer.src_lang = target_lang  # Assuming the input is always in English
+    tokenizer.src_lang = target_lang 
     encoded_text = tokenizer(text, return_tensors="pt")
     generated_tokens = model.generate(**encoded_text)
     translated_text = tokenizer.batch_decode(generated_tokens, skip_special_tokens=True)[0]
@@ -33,6 +33,7 @@ def translate_speech(audio, target_lang):
 
     return audio_path
 
+
 def translate_speech_interface(audio, target_lang):
     translated_audio = translate_speech(audio, target_lang)
     translated_audio_bytes = open(translated_audio, "rb").read()