Update app.py

app.py

@@ -1,53 +1,58 @@
 import soundfile as sf
 import torch
-from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
-import argparse
-from glob import glob
-import torchaudio
-import subprocess
+from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor,Wav2Vec2ProcessorWithLM
 import gradio as gr
+import sox
+import subprocess
 
-resampler = torchaudio.transforms.Resample(48_000, 16_000)
 
-def get_filename(wav_file):
-    filename_local = wav_file.split('/')[-1][:-4]
-    filename_new = '/tmp/'+filename_local+'_16.wav'
-    
+def read_file_and_process(wav_file):
+    filename = wav_file.split('.')[0]
+    filename_16k = filename + "16k.wav"
+    resampler(wav_file, filename_16k)
+    speech, _ = sf.read(filename_16k)
+    inputs = processor(speech, sampling_rate=16_000, return_tensors="pt", padding=True)
     
-    subprocess.call(["sox {} -r {} -b 16 -c 1 {}".format(wav_file, str(16000), filename_new)], shell=True)
-    return filename_new
+    return inputs
 
-def parse_transcription(wav_file):
-    # load pretrained model
-    processor = Wav2Vec2Processor.from_pretrained("Harveenchadha/vakyansh-wav2vec2-hindi-him-4200")
-    model = Wav2Vec2ForCTC.from_pretrained("Harveenchadha/vakyansh-wav2vec2-hindi-him-4200")
 
-    # load audio
+def resampler(input_file_path, output_file_path):
+    command = (
+        f"ffmpeg -hide_banner -loglevel panic -i {input_file_path} -ar 16000 -ac 1 -bits_per_raw_sample 16 -vn "
+        f"{output_file_path}"
+    )
+    subprocess.call(command, shell=True)
 
-    
-    wav_file = get_filename(wav_file.name)
-    audio_input, sample_rate = sf.read(wav_file)
-    #test_file = resampler(test_file[0])
 
-    # pad input values and return pt tensor
-    input_values = processor(audio_input, sampling_rate=16_000, return_tensors="pt").input_values
 
-    # INFERENCE
-    # retrieve logits & take argmax
-    logits = model(input_values).logits
+def parse_transcription(logits):
     predicted_ids = torch.argmax(logits, dim=-1)
-
-    # transcribe
     transcription = processor.decode(predicted_ids[0], skip_special_tokens=True)
     return transcription
 
+def parse(wav_file):
+    input_values = read_file_and_process(wav_file)
+    with torch.no_grad():
+        logits = model(**input_values).logits
+   
+
+    
+model_id = "Harveenchadha/vakyansh-wav2vec2-hindi-him-4200"
+processor = Wav2Vec2Processor.from_pretrained(model_id)
+processor_with_LM = Wav2Vec2ProcessorWithLM.from_pretrained(model_id)
+model = Wav2Vec2ForCTC.from_pretrained(model_id)
+
+    
+input_ = gr.Audio(source="microphone", type="filepath") 
+txtbox = gr.Textbox(
+            label="Hindi text output:",
+            lines=5
+        )
 
 title = "Speech-to-Text (Hindi) using Vakyansh"
 description = "Upload a hindi audio clip, and let AI do the hard work of transcribing."
 article = "<p style='text-align: center'><a href='https://arxiv.org/abs/2104.06678'>Large-Scale Self- and Semi-Supervised Learning for Speech Translation</a></p>"
-gr.Interface( 
-    parse_transcription,
-    title=title,
-    inputs=gr.inputs.Audio(label="Record Audio File", type="filepath", source = "microphone"),
-    description=description, article = article, outputs = "text").launch(inline = False)
 
+gr.Interface(parse, inputs = input_,  outputs=txtbox, title=title, description=description, article = article,
+             streaming=True, interactive=True,
+             analytics_enabled=False, show_tips=False, enable_queue=True).launch(inline=False);