Parakeet TDT v3 โ€” ONNX INT8

ONNX INT8 conversion of NVIDIA Parakeet-TDT 0.6B v2 for Android and Linux deployment via ONNX Runtime. Supports 114 languages with TDT greedy decoder.

Files

File Description
parakeet-encoder-int8.onnx FastConformer encoder, INT8 quantized
parakeet-decoder-joint-int8.onnx LSTM decoder + TDT joint network
vocab.json SentencePiece vocabulary (1024 tokens)
config.json Model configuration

Usage

Used by speech-android:

val asr = ParakeetASR.create(context)
val text = asr.transcribe(audioSamples, sampleRate = 16000)

Downloads last month
33
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for aufklarer/Parakeet-TDT-v3-ONNX

Quantized
(6)
this model