Update app.py

app.py

@@ -3,8 +3,6 @@
 from transformers import pipeline
 #, Wav2Vec2CTCTokenizer, Wav2Vec2ForCTC, Wav2Vec2ProcessorWithLM, WhisperFeatureExtractor, WhisperProcessor, WhisperModel, WhisperTokenizer
 import torch
-#import pyctcdecode
-#import kenlm
 import gradio as gr
 import librosa
 import os
@@ -12,15 +10,9 @@ import time
 
 #Loading the model and the tokenizer
 token_key = os.environ.get("HUGGING_FACE_HUB_TOKEN")
-#model_name = "unilux/wav2vec-xls-r-Luxembourgish20-with-LM"
-#model_name = "unilux/wav2vec-xlsr-300m-Luxembourgish-with-LM"
 
 model_name = "pgilles/whisper-large-v2-lb_cased_01"
 
-#tokenizer = Wav2Vec2CTCTokenizer.from_pretrained(model_name, use_auth_token=token_key)
-#model = Wav2Vec2ForCTC.from_pretrained(model_name, use_auth_token=token_key)
-#processor = Wav2Vec2ProcessorWithLM.from_pretrained(model_name, use_auth_token=token_key)
-
 #tokenizer = WhisperTokenizer.from_pretrained(model_name)
 #model = WhisperModel.from_pretrained(model_name)
 #processor = WhisperProcessor.from_pretrained(model_name)
@@ -28,12 +20,6 @@ model_name = "pgilles/whisper-large-v2-lb_cased_01"
 #p = pipeline("automatic-speech-recognition", model=model, tokenizer=tokenizer, feature_extractor=processor.feature_extractor, decoder=processor.decoder, use_auth_token=token_key)
 p = pipeline("automatic-speech-recognition", model=model_name, device=0, use_auth_token=token_key)
 
-#p = pipeline("automatic-speech-recognition", model=model_name, use_auth_token = token_key)
-#p = pipeline("automatic-speech-recognition", model=model_name, use_auth_token = True)
-
-#tokenizer = Wav2Vec2Tokenizer.from_pretrained(model_name)
-#model = Wav2Vec2ForCTC.from_pretrained(model_name)
-
 def load_data(input_file):
   
   """ Function for resampling to ensure that the speech input is sampled at 16KHz.
@@ -44,22 +30,21 @@ def load_data(input_file):
   #speech = librosa.effects.trim(speech, top_db= 10)
   return speech
     
-def asr_pipe(input_file):
-  load_data(input_file)
-  transcription = p(input_file, chunk_length_s=29)["text"]
+def asr_pipe(input_file, input_file_microphone, chunks):
+  input_file = input_file_microphone if input_file_microphone else input_file
+  transcription = p(input_file, chunk_length_s= chunks)["text"]
   return transcription
 
-#input1=gr.inputs.Audio(source="microphone", type='filepath', label="Click and Start Speaking...")
-#input2=gr.inputs.Audio(source="upload", type='filepath', label="Load your own file...")
-#inputs=[input1, input2]
 
-gr.Interface(asr_pipe,
-             #inputs = gr.inputs.Audio(source="microphone", type="filepath", optional=True, label="Hei kënnt Dir Är Sprooch iwwert de Mikro ophuelen"),
-             inputs=gr.inputs.Audio(source="upload", type='filepath', label="Load your own file..."),
-             #inputs = inputs,
+gr.Interface(fn = asr_pipe,
+             inputs=[
+                 gr.inputs.Audio(source="upload", type='filepath', label="Load your own file...", optional = True),
+                 gr.inputs.Audio(source="microphone", type="filepath", label="Record from microphone", optional = True),
+                 gr.Slider(minimum=3, maximum=32, value=30, step=1, label="Chunk Length")
+             ]
              outputs = gr.outputs.Textbox(label="Erkannten Text"),
              title="Sproocherkennung fir d'Lëtzebuergescht @uni.lu",
-             description = "Dës App convertéiert Är geschwate Sprooch an de (méi oder manner richtegen ;-)) Text! Dir kënnt Iech selwer iwwer de Mikro ophuelen (am beschten 5 bis 10 Sekonnen), eng Datei eroplueden oder e Beispill ënnen auswielen.",
+             description = "Dës App convertéiert Är geschwate Sprooch an de (méi oder manner richtegen ;-)) Text! Dir kënnt Iech selwer iwwer de Mikro ophuelen, eng Datei eroplueden oder e Beispill ënnen auswielen.",
              examples = [["Chamber2022_1.wav"], ["Chamber2022_2.wav"], ["Chamber2022_3.wav"], ["Chamber2022_4.wav"]], 
              theme="default").launch()