Spaces:
Sleeping
Sleeping
File size: 3,396 Bytes
c0e8dca 0c33dd3 c0e8dca 5928b3c 0c33dd3 414f828 64dc005 414f828 64dc005 c0e8dca 0c33dd3 c0e8dca 0c33dd3 c0e8dca 0c33dd3 c0e8dca 0c33dd3 c0e8dca 0c33dd3 c0e8dca 0c33dd3 c0e8dca 0c33dd3 c0e8dca 1f8f25c c0e8dca |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 |
# app.py
import gradio as gr
from transformers import pipeline
import numpy as np
import librosa # pip install librosa
# --- EDIT THIS: map display names to your HF Hub model IDs ---
language_models = {
"Akan (Asante Twi)": "FarmerlineML/w2v-bert-2.0_twi_alpha_v1",
"Ewe": "FarmerlineML/w2v-bert-2.0_ewe_2",
"Kiswahili": "FarmerlineML/w2v-bert-2.0_swahili_alpha",
"Luganda": "FarmerlineML/w2v-bert-2.0_luganda",
"Brazilian Portuguese": "FarmerlineML/w2v-bert-2.0_brazilian_portugese_alpha",
"Fante": "misterkissi/w2v2-lg-xls-r-300m-fante",
"Bemba": "DarliAI/kissi-w2v2-lg-xls-r-300m-bemba",
"Bambara": "DarliAI/kissi-w2v2-lg-xls-r-300m-bambara",
"Dagaare": "DarliAI/kissi-w2v2-lg-xls-r-300m-dagaare",
"Kinyarwanda": "DarliAI/kissi-w2v2-lg-xls-r-300m-kinyarwanda",
"Fula": "DarliAI/kissi-wav2vec2-fula-fleurs-full",
"Oromo": "DarliAI/kissi-w2v-bert-2.0-oromo",
"Runyankore": "misterkissi/w2v2-lg-xls-r-300m-runyankore",
"Ga": "misterkissi/w2v2-lg-xls-r-300m-ga",
"Vai": "misterkissi/whisper-small-vai",
"Yoruba": "FarmerlineML/w2v-bert-2.0_yoruba_v1",
"Luganda": "FarmerlineML/luganda_fkd",
"Luo": "FarmerlineML/w2v-bert-2.0_luo_v2",
"Igbo": "FarmerlineML/w2v-bert-2.0_igbo_v1"
# add more as needed
}
# Pre-load pipelines for each language on CPU (device=-1)
asr_pipelines = {
lang: pipeline(
task="automatic-speech-recognition",
model=model_id,
device=-1, # force CPU usage
chunk_length_s=30
)
for lang, model_id in language_models.items()
}
def transcribe(audio_path: str, language: str) -> str:
"""
Load the audio via librosa (supports mp3, wav, flac, m4a, ogg, etc.),
convert to mono, then run it through the chosen ASR pipeline.
"""
if not audio_path:
return "⚠️ Please upload or record an audio clip."
# librosa.load returns a 1D np.ndarray (mono) and the sample rate
speech, sr = librosa.load(audio_path, sr=None, mono=True)
# Call the Hugging Face ASR pipeline
result = asr_pipelines[language]({
"sampling_rate": sr,
"raw": speech
})
return result.get("text", "")
with gr.Blocks(title="🌐 Multilingual ASR Demo") as demo:
gr.Markdown(
"""
## 🎙️ Multilingual Speech-to-Text
Upload an audio file (MP3, WAV, FLAC, M4A, OGG,…) or record via your microphone.
Then choose the language/model and hit **Transcribe**.
"""
)
with gr.Row():
lang = gr.Dropdown(
choices=list(language_models.keys()),
value=list(language_models.keys())[0],
label="Select Language / Model"
)
with gr.Row():
audio = gr.Audio(
sources=["upload", "microphone"],
type="filepath",
label="Upload or Record Audio"
)
btn = gr.Button("Transcribe")
output = gr.Textbox(label="Transcription")
btn.click(fn=transcribe, inputs=[audio, lang], outputs=output)
if __name__ == "__main__":
demo.launch()
|