Update app.py

app.py

@@ -22,8 +22,15 @@ Tonic's Unity On Device!🚀 on your own data & in your own way by cloning
 TeamTonic is always making cool demos! Join our active builder's community on Discord: [Discord](https://discord.gg/GWpVpekp) On 🤗Huggingface: [TeamTonic](https://huggingface.co/TeamTonic) & [MultiTransformer](https://huggingface.co/MultiTransformer) On 🌐Github: [Polytonic](https://github.com/tonic-ai) & contribute to 🌟 [PolyGPT](https://github.com/tonic-ai/polygpt-alpha)"             
 """
 
+def save_and_resample_audio(input_audio_path, output_audio_path, resample_rate=16000):
+    waveform, sample_rate = torchaudio.load(input_audio_path)
 
-def save_audio(audio_input, output_dir="saved_audio"):
+    resampler = torchaudio.transforms.Resample(sample_rate, resample_rate, dtype=waveform.dtype)
+    resampled_waveform = resampler(waveform)
+
+    torchaudio.save(output_audio_path, resampled_waveform, resample_rate)
+
+def save_audio(audio_input, output_dir="saved_audio", resample_rate=16000):
     if not os.path.exists(output_dir):
         os.makedirs(output_dir)
     sample_rate, audio_data = audio_input
@@ -31,12 +38,15 @@ def save_audio(audio_input, output_dir="saved_audio"):
     file_path = os.path.join(output_dir, file_name)
     sf.write(file_path, audio_data, sample_rate)
 
-    return file_path
+    resampled_file_path = os.path.join(output_dir, f"resampled_{file_name}")
+    save_and_resample_audio(file_path, resampled_file_path, resample_rate)
+
+    return resampled_file_path
 
 def speech_to_text(audio_data, tgt_lang):
     file_path = save_audio(audio_data)
     audio_input, _ = torchaudio.load(file_path)
-    s2t_model = torch.jit.load("unity_on_device.ptl")
+    s2t_model = torch.jit.load("unity_on_device.ptl", map_location=torch.device('cpu'))
     with torch.no_grad():
         model_output = s2t_model(audio_input, tgt_lang=languages[tgt_lang])
     transcribed_text = model_output[0] if model_output else ""
@@ -47,7 +57,7 @@ def speech_to_text(audio_data, tgt_lang):
 def speech_to_speech_translation(audio_data, tgt_lang):
     file_path = save_audio(audio_data)
     audio_input, _ = torchaudio.load(file_path)
-    s2st_model = torch.jit.load("unity_on_device.ptl")
+    s2st_model = torch.jit.load("unity_on_device.ptl", map_location=torch.device('cpu'))
     with torch.no_grad():
         translated_text, units, waveform = s2st_model(audio_input, tgt_lang=languages[tgt_lang])
     output_file = "/tmp/result.wav"