sugar

nvidia/nemotron-3.5-asr-streaming-0.6b fine-tuned for Akan (Twi), with replay of English, French, Spanish, German and Portuguese so the original languages are kept. New language prompts: ak-GH. All original prompts (and auto) still work.

The sugar.nemo file at the root is the best checkpoint; final/sugar.nemo is the model at the last training step. This card describes the best one: step 18000 of run beaebfcbbf53 - selection score 55.806 (target WER + 2.0 x average replay WER increase over the base model).

Language Prompt WER base WER this model
Akan (Twi) ak-GH new 87.29 54.71
English en-US kept 3.53 3.84
French fr-FR kept 8.76 9.3
Spanish es-ES kept 5.74 6.48
German de-DE kept 8.66 8.59
Portuguese pt-BR kept 8.22 9.37

WER is on held-out clips, lower-cased, punctuation and language tags removed. "auto" language-ID accuracy on the new languages: 75.5 %.

Use (NeMo)

import nemo.collections.asr as nemo_asr
from huggingface_hub import hf_hub_download

model = nemo_asr.models.ASRModel.restore_from(hf_hub_download("prince4332/sugar", "sugar.nemo"))
print(model.transcribe(["clip.wav"], target_lang="ak-GH")[0].text)   # or "en-US", "auto", ...

Streaming: model.set_inference_prompt("ak-GH"), then the usual cache-aware loop (CacheAwareStreamingAudioBuffer + conformer_stream_step). Language tags like <ak-GH> are stripped by the decoder (decoding.strip_lang_tags).

Downloads last month
358
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for prince4332/sugar

Finetuned
(63)
this model