Live Demo Khasi Garo Pnar Hindi English

Hear it. Read it.

Type something to hear it spoken in Khasi, Garo, Pnar, Hindi or English, or upload Khasi audio to have it transcribed.

Preview mode. The interface is running against a local stub — no model is being called yet.

Results are machine-generated and may contain mistakes.

Text

0 / 400

Voice

Language

Advanced parameters

Inference steps 32

More steps = cleaner audio, slower to generate.

Guidance scale 3.0

Higher = closer to your text, but can sound stiff.

Speed 0.9

How fast the voice talks.

Preprocess prompt Postprocess output

Output

Generated speech will appear here.

Audio

Transcription currently supports Khasi audio only.

Drop an audio file, or browse MP3 or WAV · up to 30 seconds

Output format

Continuous text Timestamps Speaker identification

Advanced parameters

Enable VAD filter

Skips silence so the model doesn't guess words there.

Repetition penalty 1.2

Higher = fewer repeated words, but may skip some entirely.

Beam size 5

Higher = more accurate search, but slower to transcribe.

VAD min silence (ms) 1200

How long a pause must be before it's treated as a break.

No-speech threshold 0.9

How sure the model must be that a part is silent before skipping it.

Transcript

The transcript will appear here.

Models run on Tynrai's own Hugging Face Spaces. See the models