Create new file

app.py

@@ -0,0 +1,42 @@
+import librosa
+import gradio as gr
+#from transformers import Wav2Vec2Tokenizer, Wav2Vec2ForCTC
+from transformers import pipeline
+
+#Loading the model and the tokenizer
+model_name = "unilux/wav2vec-xls-r-Luxembourgish20-with-LM"
+pipe = pipeline("automatic-speech-recognition", model=model_name)
+
+#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.
+ """
+ #read the file
+ speech, sample_rate = librosa.load(input_file)
+ #make it 1-D
+ if len(speech.shape) > 1: 
+ speech = speech[:,0] + speech[:,1]
+ #Resampling at 16KHz since wav2vec2-base-960h is pretrained and fine-tuned on speech audio sampled at 16 KHz.
+ if sample_rate !=16000:
+ speech = librosa.resample(speech, sample_rate,16000)
+ return speech
+ 
+def asr_pipe(input_file):
+ transcription = pipe(input_file, chunk_length_s=3, stride_length_s=(0.5, 0.5))
+ return transcription
+ 
+
+gr.Interface(asr_pipe,
+ inputs = [
+ gr.inputs.Audio(source="microphone", type="filepath", optional=True, label="Hei kënnt Dir Är Sprooch iwwert de Mikro ophuelen"),
+ gr.inputs.Audio(source="upload", type='filepath', optional=True)
+ ],
+ outputs = gr.outputs.Textbox(label="Output Text"),
+ title="Sproocherkennung fir d'Lëtzebuergescht @uni.lu",
+ description = "Dës App convertéiert Är geschwate Sprooch an de (méi oder manner richegen ;-)) Text!",
+ examples = [["ChamberMeisch.wav"], ["Chamber_Fayot_2005.wav"]], theme="default").launch()
+