up

README.md

@@ -1,8 +1,8 @@
 ---
-title: XLS R 2B 21 EN
-emoji: 🏢
-colorFrom: green
-colorTo: indigo
+title: XLS R 2B 22 EN
+emoji: 📊
+colorFrom: gray
+colorTo: red
 sdk: gradio
 app_file: app.py
 pinned: false

app.py

@@ -0,0 +1,37 @@
+import gradio as gr
+import librosa
+from transformers import AutoFeatureExtractor, AutoTokenizer, SpeechEncoderDecoderModel
+
+model_name = "facebook/wav2vec2-xls-r-2b-21-to-en",
+    
+feature_extractor = AutoFeatureExtractor.from_pretrained(model_name)
+tokenizer = AutoTokenizer.from_pretrained(model_name, use_fast=False)
+model = SpeechEncoderDecoderModel.from_pretrained(model_name)
+
+def process_audio_file(file):
+    data, sr = librosa.load(file)
+    if sr != 16000:
+        data = librosa.resample(data, sr, 16000)
+    input_values = feature_extractor(data, return_tensors="pt").input_values
+    return input_values
+    
+def transcribe(file):
+    input_values = process_audio_file(file)
+    
+    sequences = model.generate(input_values, num_beams=1, max_length=30)
+    
+    transcription = tokenizer.batch_decode(sequences, skip_special_tokens=True)
+    return transcription[0]
+
+iface = gr.Interface(
+    fn=transcribe, 
+    inputs=[
+        gr.inputs.Audio(source="microphone", type='filepath'),
+    ],
+    outputs="text",
+    layout="horizontal",
+    theme="huggingface",
+    title="XLS-R 2B 21-to-EN Speech Translation",
+    description="A simple interface to translate from 21 spoken languages to written English.",
+)
+iface.launch()

packages.txt

@@ -0,0 +1,2 @@
+libsndfile1
+sox

requirements.txt

@@ -0,0 +1,5 @@
+SoundFile==0.9.0.post1
+librosa
+sentencepiece
+torch
+git+git://github.com/huggingface/transformers