OpenThai2.0-ASR-Streaming GGUF

GGUF build of iapp/OpenThai2.0-ASR-Streaming for the audio.cpp nemotron_asr family. Converted from openthai2.0-asr-streaming_bias-1.0.nemo. Weights were not retrained.

The source is iApp Technology's Thai fine-tune of nvidia/nemotron-3.5-asr-streaming-0.6b. It is released under OpenMDW-1.1; see LICENSE.

File Type Size
openthai2.0-asr-streaming-f16.gguf F16 1.2 GB
openthai2.0-asr-streaming-q8_0.gguf Q8_0 891 MB
openthai2.0-asr-streaming-q6_k.gguf Q6_K 802 MB
openthai2.0-asr-streaming-q5_k.gguf Q5_K 734 MB
openthai2.0-asr-streaming-q4_k.gguf Q4_K 670 MB
openthai2.0-asr-streaming-q3_k.gguf Q3_K 602 MB
openthai2.0-asr-streaming-q2_k.gguf Q2_K 550 MB
openthai2.0-asr-streaming-tq2_1.gguf TQ2_1 518 MB

On a short test clip at lookahead 6, F16, Q8_0, Q6_K, Q5_K, Q4_K, and Q3_K match the NeMo transcript. Q2_K and TQ2_1 do not; check them on your own audio before use.

Run

The file's default language is th-TH. Supported lookahead values are 0, 3, 6, and 13. A lookahead of 6 is 560 ms, the same setting as the NeMo notebook.

audiocpp_cli --task asr --family nemotron_asr \
  --model openthai2.0-asr-streaming-q8_0.gguf \
  --backend cpu --mode streaming --audio speech_16k.wav \
  --language th-TH --request-option lookahead_tokens=6 \
  --text-out transcript.txt

th-TH writes everything in Thai script. th-mix keeps English words in Latin script. Input audio is 16 kHz mono.

Downloads last month
384
GGUF
Model size
0.6B params
Architecture
audiocpp
Hardware compatibility
Log In to add your hardware

2-bit

6-bit

8-bit

16-bit

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

Model tree for FILM6912/OpenThai2.0-ASR-Streaming-GGUF

Quantized
(1)
this model