Skip to content
Audio

Soriya-ASR-Khmer-325M

компактное распознавание кхмерской речи

1 min read

Модель автоматического распознавания речи для кхмерского языка (km) от Nukthean. Она объединяет 12-слойную Conformer-аудиобашню Google EmbeddingGemma-2 и 6-слойный Transformer CTC Reader. Около 324.8M параметров, инференс-файл model.pt — примерно 652 MB; лицензия Apache 2.0.

Ключевые особенности

  • CTC-ASR только для кхмерского, словарь на 582 субтокена.
  • Архитектура: Conformer-аудиобашня EmbeddingGemma-2 (d=1024) + 6-слойный Transformer CTC Reader (d=512, 8 голов).
  • Компактный размер: ~324.8M п��раметров и ~652 MB для инференса.
  • Длинные записи автоматически нарезаются на окна по 30 секунд.
  • Обучение: сначала на Digital Divide Data, затем joint multi-domain с замороженным Conformer и weight soup Reader (α=0.60).
  • Ограничения: без пунктуации, числа — кхмерскими словами, английские слова — фонетически кхмерским письмом.

Пример

import torch
from soriya_asr import load_model, transcribe

device = "cuda" if torch.cuda.is_available() else "cpu"
model, units = load_model("Nukthean/soriya-asr-khmer-325m", device=device)
text = transcribe(model, units, "sample.wav")
print(text)
ПодробнееCER 3.91% и WER 10.31% на стратифицированной выборке Digital Divide Data (8.60 ч, 3648 клипов); exact match 29.6%.
На unseen FLEURS test CER 15.85%, WER 29.28% (khmercut), exact match 3.2%.
Обучение шло на двух Tesla T4 (2×15 GB VRAM) в Kaggle; ин��еренс тестировали на CPU в PyTorch FP32.
Нужен 16 kHz mono; пунктуация не генерируется, числа выводятся как произнесённые кхмерские слова, английские слова — кхмерской фонетикой.
Discuss Telegram 1

Will it run?

Working out the memory needs

The bot is still collecting this model's weight sizes; the estimate will appear here in a few minutes. Meanwhile you can rent a GPU with headroom.

Massed ComputeOpen Massed Compute

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # Nukthean/soriya-asr-khmer-325m
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download Nukthean/soriya-asr-khmer-325m --local-dir ~/models/soriya-asr-khmer-325m
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to soriya-asr-khmer-325m
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of soriya-asr-khmer-325m (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

Nukthean/soriya-asr-khmer-325m
Author
Nukthean
Category
Audio
Released
Posted

On NeyroLearn

Hugging Face

01

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments