up

README.md

@@ -1,8 +1,8 @@
 ---
-title: XLS R 2B 22 16
-emoji: 💻
-colorFrom: green
-colorTo: purple
+title: XLS R 300m EN 15
+emoji: 📊
+colorFrom: gray
+colorTo: red
 sdk: gradio
 app_file: app.py
 pinned: false

app.py

@@ -0,0 +1,81 @@
+import gradio as gr
+import librosa
+from transformers import AutoFeatureExtractor, AutoTokenizer, SpeechEncoderDecoderModel
+
+model_name = "facebook/wav2vec2-xls-r-300m-22-to-16",
+    
+feature_extractor = AutoFeatureExtractor.from_pretrained(model_name, use_auth_token="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")
+tokenizer = AutoTokenizer.from_pretrained(model_name, use_auth_token="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", use_fast=False)
+model = SpeechEncoderDecoderModel.from_pretrained(model_name, use_auth_token="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")
+
+def process_audio_file(file):
+    data, sr = librosa.load(file)
+    if sr != 16000:
+        data = librosa.resample(data, sr, 16000)
+    print(data.shape)
+    input_values = feature_extractor(data, return_tensors="pt").input_values
+    return input_values
+    
+def transcribe(file, target_language):
+    
+    target_code = target_language.split("(")[-1].split(")")[0]
+    forced_bos_token_id = MAPPING[target_code]
+
+    input_values = process_audio_file(file)
+    
+    sequences = model.generate(input_values, forced_bos_token_id=forced_bos_token_id)
+    
+    transcription = tokenizer.batch_decode(sequences, skip_special_tokens=True)
+    return transcription[0]
+    
+target_language = [
+    "English (en)",
+    "German (de)",
+    "Turkish (tr)",
+    "Persian (fa)",
+    "Swedish (sv)",
+    "Mongolian (mn)",
+    "Chinese (zh)",
+    "Welsh (cy)",
+    "Catalan (ca)",
+    "Slovenian (sl)",
+    "Estonian (et)",
+    "Indonesian (id)",
+    "Arabic (ar)",
+    "Tamil (ta)",
+    "Latvian (lv)",
+    "Japanese (ja)",
+]
+
+MAPPING = {
+    "en": 250004,
+    "de": 250003,
+    "tr": 250023,
+    "fa": 250029,
+    "sv": 250042,
+    "mn": 250037,
+    "zh": 250025,
+    "cy": 250007,
+    "ca": 250005,
+    "sl": 250052,
+    "et": 250006,
+    "id": 250032,
+    "ar": 250001,
+    "ta": 250044,
+    "lv": 250017,
+    "ja": 250012,
+}
+    
+iface = gr.Interface(
+    fn=transcribe, 
+    inputs=[
+        gr.inputs.Audio(source="microphone", type='filepath'),
+        gr.inputs.Dropdown(target_language),
+    ],
+    outputs="text",
+    layout="horizontal",
+    theme="huggingface",
+    title="XLS-R 300M 22-to-16 Speech Translation",
+    description="A simple interface to translate from 22 input spoken languages to 16 written languages.",
+)
+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