Skip to content
Audio

Whisper Large-v3 + чилийский LoRA

ASR для телефонных колл-центров

1 min readupdated

Это LoRA-адаптер к openai/whisper-large-v3 от HablaIP SpA (проект CORFO 25CH2-306482). Он настроен на спонтанную чилийскую испанскую речь в реальных телефонных разговорах колл-центра: местные выражения, шум линии и живая речь. Обучается примерно 1% параметров базовой модели, остальные веса заморожены.

Ключевые особенности

  • WER 21.18% против 34.76% у базового Whisper Large-v3 — относительное снижение ошибки на 39%.
  • CER 11.37%; разница статистически значима по bootstrap с группировкой по звонкам.
  • LoRA r=32, alpha=64, dropout=0.05 на q_proj/v_proj в блоках внимания.
  • Обучение: около 1 800 клипов (~8 часов) реального аудио колл-центра, ручная пословная транскрипция.
  • Данные приватны; разбиение по целым звонкам, чтобы избежать утечки фрагментов.
  • Ограничение: телефонная чилийская речь колл-центра; другие домены и акценты не охарактеризованы.

Пример

from transformers import WhisperProcessor, WhisperForConditionalGeneration
from peft import PeftModel
import torch
processor = WhisperProcessor.from_pretrained("openai/whisper-large-v3", language="spanish", task="transcribe")
model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-large-v3", torch_dtype=torch.bfloat16)
model = PeftModel.from_pretrained(model, "hablaip/whisper-large-v3-chileno-lora").eval()
inputs = processor.feature_extractor(audio, sampling_rate=16000, return_tensors="pt")
ПодробнееWER: 34.76% → 21.18%, CER: 11.37%, относительное снижение ошибки: 39%.
LoRA r=32, alpha=64, dropout=0.05; обучалось ~1% параметров.
Данные: ~1 800 клипов (~8 часов), приватные, транскрипция verbatim.
Ограничение: телефонная чилийская речь колл-центра; другие домены и акценты не охарактеризованы.
Discuss Telegram 5 21

Will it run?

Runs in F16

Fits without quantisation.

  • F16weights 2.9 GB · ready files3.7 GB
  • FP8 / INT8recommendedweights 1.5 GB2.1 GB
  • 4-bitweights 0.9 GB1.5 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · F16 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

This is a LoRA: counted together with its base model whisper-large-v3.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

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.

    # hablaip/whisper-large-v3-chileno-lora
    # ===== 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 openai/whisper-large-v3 --local-dir ~/models/whisper-large-v3
    hf download hablaip/whisper-large-v3-chileno-lora --local-dir ~/models/whisper-large-v3-chileno-lora
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to whisper-large-v3
  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

Requires the base model

openai/whisper-large-v3

This model is a LoRA / adapter. Download the base model first, then load this one on top.

New posts about versions of whisper-large-v3 (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 3All on Hugging Face

About the model

hablaip/whisper-large-v3-chileno-lora
Author
hablaip
Category
Audio
Released
Posted

On NeyroLearn

Hugging Face

215

Ask AI about this model

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

Comments