derkar00/mms-tuk-latin-asr

MMS Turkmen ASR (Latin Script)

This model performs Automatic Speech Recognition (ASR) for the Turkmen language using the Latin script (tuk-script_latin).

It is based on facebook/mms-1b-all, Meta's Massively Multilingual Speech model covering 1000+ languages.

Usage

import librosa
import torch
from transformers import Wav2Vec2ForCTC, AutoProcessor

processor = AutoProcessor.from_pretrained("derkar00/mms-tuk-latin-asr")
model = Wav2Vec2ForCTC.from_pretrained("derkar00/mms-tuk-latin-asr")

processor.tokenizer.set_target_lang("tuk-script_latin")
model.load_adapter("tuk-script_latin")

# Load audio (must be 16kHz mono)
audio, rate = librosa.load("your_audio.wav", sr=16000, mono=True)

inputs = processor(audio, sampling_rate=16000, return_tensors="pt")
with torch.no_grad():
    logits = model(**inputs).logits

predicted_ids = torch.argmax(logits, dim=-1)
print(processor.decode(predicted_ids[0]))

Language

• Language: Turkmen (tuk)
• Script: Latin
• Task: Automatic Speech Recognition

Limitations

• Accuracy may vary on noisy audio or strong accents
• Turkmen is a low-resource language so training data is limited
• Works best on clean, clear speech

Base Model

• facebook/mms-1b-all
• Paper: Scaling Speech Technology to 1000+ Languages