return to old PPM model

app.py

@@ -23,9 +23,9 @@ transformation = jiwer.Compose([
 
 # WPM part
 from transformers import Wav2Vec2Processor, Wav2Vec2ForCTC
-processor = Wav2Vec2Processor.from_pretrained("vitouphy/wav2vec2-xls-r-300m-timit-phoneme")
-model = Wav2Vec2ForCTC.from_pretrained("vitouphy/wav2vec2-xls-r-300m-timit-phoneme")
-phoneme_model =  pipeline(model="vitouphy/wav2vec2-xls-r-300m-timit-phoneme") 
+processor = Wav2Vec2Processor.from_pretrained("facebook/wav2vec2-xlsr-53-espeak-cv-ft")
+phoneme_model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-xlsr-53-espeak-cv-ft")
+# phoneme_model =  pipeline(model="vitouphy/wav2vec2-xls-r-300m-timit-phoneme") 
 class ChangeSampleRate(nn.Module):
     def __init__(self, input_rate: int, output_rate: int):
         super().__init__()
@@ -83,9 +83,6 @@ def calc_mos(audio_path, ref):
         logits = phoneme_model(out_wavs).logits
     phone_predicted_ids = torch.argmax(logits, dim=-1)
     phone_transcription = processor.batch_decode(phone_predicted_ids)
-    
-    # Disable PPM for now
-    phone_transcription = ['D U M M Y']
     lst_phonemes = phone_transcription[0].split(" ")
     wav_vad = torchaudio.functional.vad(wav, sample_rate=sr)
 
@@ -94,7 +91,6 @@ def calc_mos(audio_path, ref):
 
     ppm = len(lst_phonemes) / (wav_vad.shape[-1] / sr) * 60
     
-    # pdb.set_trace()
     return AVA_MOS, MOS_fig, INTELI_score, INT_fig, trans, phone_transcription, ppm, f0_db_fig