Upload 36 files

.gitattributes

@@ -33,3 +33,9 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+Samples/emp2.wav filter=lfs diff=lfs merge=lfs -text
+Samples/Montreal[[:space:]]Vacation[[:space:]]Travel[[:space:]]Guide[[:space:]][[:space:]]Expedia.mp3 filter=lfs diff=lfs merge=lfs -text
+Samples/Que[[:space:]]es[[:space:]]TED[[:space:]]y[[:space:]]TEDx.mp3 filter=lfs diff=lfs merge=lfs -text
+Samples/test_mixture.wav filter=lfs diff=lfs merge=lfs -text
+Samples/Gujarati_2.wav filter=lfs diff=lfs merge=lfs -text
+Samples/Telugu_2.wav filter=lfs diff=lfs merge=lfs -text

README.md

@@ -1,13 +1,12 @@
 ---
 title: Meta Mms ASR
-emoji: 👀
-colorFrom: blue
-colorTo: green
+emoji: 📚
+colorFrom: yellow
+colorTo: pink
 sdk: gradio
-sdk_version: 4.1.1
+sdk_version: 4.0.2
 app_file: app.py
 pinned: false
-license: mit
 ---
 
 Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

Samples/Gujarati_2.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3d087ac4bc15ddccecbbded4c8fdd2c9501bd8e08c6550f9c03ec46bd2df64da
+size 1597484

Samples/Montreal Vacation Travel Guide Expedia.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bc5148bfd0aca841584c6120c9b02c4c3d9aada2e9b015827a7869803b6471cb
+size 8058808

Samples/Que es TED y TEDx.mp3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:120a8f591bec8715dea618a14bf0538ec03444aaf875994bac0d8a1506f2ffdc
+size 1852114

Samples/Telugu_2.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ccee4c1d338d4d6fc96dfca5f48ecac3d9b9517e44f0302037fe46304e9e76b0
+size 1122348

Samples/emp2.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:133ac69ac8a7d99f1fe92b7dbd810d7f4e44cd4f282179aa8c5d83183e64ad61
+size 1122066

Samples/test_mixture.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7126269f058724336d28bf6e82fa14b7030de321dbafa48c9fa20884d35dab9d
+size 1546250

app.py

@@ -0,0 +1,28 @@
+import gradio as gr
+from asr import transcribe,detect_language,transcribe_lang
+
+demo = gr.Interface(transcribe,
+                   inputs = "microphone",
+                   # gr.Audio(sources=["microphone"]),
+                   outputs=["text","text"],
+                   examples=["./Samples/Hindi_1.mp3","./Samples/Hindi_2.mp3","./Samples/Tamil_1.mp3","./Samples/Tamil_2.mp3","./Samples/Marathi_1.mp3","./Samples/Marathi_2.mp3","./Samples/Nepal_1.mp3","./Samples/Nepal_2.mp3","./Samples/Telugu_1.wav","./Samples/Telugu_2.wav","./Samples/Malayalam_1.wav","./Samples/Malayalam_2.wav","./Samples/Gujarati_1.wav","./Samples/Gujarati_2.wav","./Samples/Bengali_1.wav","./Samples/Bengali_2.wav"]
+)
+demo2 = gr.Interface(detect_language,
+                     inputs = "microphone",
+                   # gr.Audio(sources=["microphone"]),
+                   outputs=["text","text"],
+                     examples=["./Samples/Hindi_1.mp3","./Samples/Hindi_2.mp3","./Samples/Tamil_1.mp3","./Samples/Tamil_2.mp3","./Samples/Marathi_1.mp3","./Samples/Marathi_2.mp3","./Samples/Nepal_1.mp3","./Samples/Nepal_2.mp3","./Samples/Telugu_1.wav","./Samples/Telugu_2.wav","./Samples/Malayalam_1.wav","./Samples/Malayalam_2.wav","./Samples/Gujarati_1.wav","./Samples/Gujarati_2.wav","./Samples/Bengali_1.wav","./Samples/Bengali_2.wav"]
+)
+demo3 = gr.Interface(transcribe_lang,
+                     inputs = ["microphone",gr.Radio([("Hindi","hin"),("Bengali","ben"),("Odia","ory"),("Gujarati","guj"),("Telugu","tel"),("Tamil","tam"),("Marathi","mar"),("English","eng")],value="hindi")],
+                   # gr.Audio(sources=["microphone"]),
+                   outputs=["text","text"],
+                    examples=[["./Samples/Hindi_1.mp3","hin"],["./Samples/Hindi_2.mp3","hin"],["./Samples/Hindi_3.mp3","hin"],["./Samples/Hindi_4.mp3","hin"],["./Samples/Hindi_5.mp3","hin"],["./Samples/Tamil_1.mp3","tam"],["./Samples/Tamil_2.mp3","tam"],["./Samples/Marathi_1.mp3","mar"],["./Samples/Marathi_2.mp3","mar"],["./Samples/Telugu_1.wav","tel"],["./Samples/Telugu_2.wav","tel"],["./Samples/Malayalam_1.wav","mal"],["./Samples/Malayalam_2.wav","mal"],["./Samples/Gujarati_1.wav","guj"],["./Samples/Gujarati_2.wav","guj"],["./Samples/Bengali_1.wav","ben"],["./Samples/Bengali_2.wav","ben"],["./Samples/climate ex short.wav","eng"],["./Samples/emp2.wav","eng"]]
+)
+                    
+
+tabbed_interface = gr.TabbedInterface([demo,demo2,demo3],["Transcribe by auto detecting language","Detect language","Transcribe by providing language"])
+
+with gr.Blocks() as asr:
+    tabbed_interface.render()
+asr.launch()

asr.py

@@ -0,0 +1,86 @@
+from transformers import Wav2Vec2ForCTC, AutoProcessor
+import torch
+from transformers import Wav2Vec2ForSequenceClassification, AutoFeatureExtractor
+import time
+import gradio as gr
+import librosa
+import numpy as np
+
+
+model_id = "facebook/mms-1b-all"
+processor = AutoProcessor.from_pretrained(model_id)
+model = Wav2Vec2ForCTC.from_pretrained(model_id)
+
+model_id_lid = "facebook/mms-lid-126"
+processor_lid = AutoFeatureExtractor.from_pretrained(model_id_lid)
+model_lid = Wav2Vec2ForSequenceClassification.from_pretrained(model_id_lid)
+
+def resample_to_16k(audio, orig_sr):
+    y_resampled = librosa.resample(y=audio, orig_sr=orig_sr, target_sr = 16000)
+    return y_resampled
+
+
+def transcribe(audio):
+    print(audio)
+    # audio = librosa.load(audio, sr=16_000, mono=True)[0]
+    # print("After loading: ",audio)
+    sr,y = audio
+    y = y.astype(np.float32)
+    y /= np.max(np.abs(y))
+    y_resampled = resample_to_16k(y, sr)
+    print("Without using librosa to load:",y_resampled)
+    # inputs = processor(audio, sampling_rate=16_000,return_tensors="pt")
+    inputs = processor(y_resampled, sampling_rate=16_000,return_tensors="pt")
+    with torch.no_grad():
+        tr_start_time = time.time()
+        outputs = model(**inputs).logits
+        tr_end_time = time.time()
+    ids = torch.argmax(outputs, dim=-1)[0]
+    transcription = processor.decode(ids)
+    return transcription,(tr_end_time-tr_start_time)
+
+
+def detect_language(audio): 
+    print(audio)
+    # audio = librosa.load(audio, sr=16_000, mono=True)[0]
+    sr,y = audio
+    y = y.astype(np.float32)
+    y /= np.max(np.abs(y))
+    y_resampled = resample_to_16k(y, sr)
+    print("Without using librosa to load:",y_resampled)
+    # inputs = processor(audio, sampling_rate=16_000,return_tensors="pt")
+    inputs = processor(y_resampled, sampling_rate=16_000,return_tensors="pt")
+    # print(audio)
+    # inputs_lid = processor_lid(audio, sampling_rate=16_000, return_tensors="pt")
+    with torch.no_grad():
+        start_time = time.time()
+        outputs_lid = model_lid(**inputs).logits
+        end_time = time.time()
+#     print(end_time-start_time," sec")
+    lang_id = torch.argmax(outputs_lid, dim=-1)[0].item()
+    detected_lang = model_lid.config.id2label[lang_id]
+    print(detected_lang)
+    return detected_lang, (end_time-start_time)
+
+
+def transcribe_lang(audio,lang):
+    # audio = librosa.load(audio, sr=16_000, mono=True)[0]
+    sr,y = audio
+    y = y.astype(np.float32)
+    y /= np.max(np.abs(y))
+    y_resampled = resample_to_16k(y, sr)
+    print("Without using librosa to load:",y_resampled)
+    processor.tokenizer.set_target_lang(lang)
+    model.load_adapter(lang)
+    print(lang)
+    # inputs = processor(audio, sampling_rate=16_000,return_tensors="pt")
+    inputs = processor(y_resampled, sampling_rate=16_000,return_tensors="pt")
+    with torch.no_grad():
+        tr_start_time = time.time()
+        outputs = model(**inputs).logits
+        tr_end_time = time.time()
+    ids = torch.argmax(outputs, dim=-1)[0]
+    transcription = processor.decode(ids)
+    return transcription,(tr_end_time-tr_start_time)
+
+

requirements.txt

@@ -0,0 +1,6 @@
+torch 
+accelerate 
+torchaudio 
+datasets
+transformers
+librosa