Skip to content
Audio

Amharic-ASR-v2

version two of the automatic speech recognition model for Amharic

2 min read

The model is built for real-world conditions and telephony, improving upon the previous version through fine-tuning on an expanded dataset (about 1,188 hours of audio) with heavy augmentation for telecom channels. The architecture uses a w2v-bert-2.0 encoder (610 million parameters) combined with a language model (KenLM 4-gram) for beam search decoding. This development helps transcribe phone calls, noisy recordings, and Amharic dialects more accurately, outperforming the first version on most test sets.

Key features

  • The average word error rate (WER) drops to 27.15% thanks to the new acoustic model and language module.
  • Improved transcription quality in harsh conditions, on narrow-band 2G connections, and with background noise.
  • Trained on a dataset of over 200,000 utterances covering various Amharic dialects.
  • Supports flexible decoding parameter tuning to balance accuracy and speed.

What you can do

  • Integrate the model into apps for automated transcription of Amharic phone conversations and calls.
  • Process long audio files using a batch pipeline in the transformers library.
  • Recognize audio recordings with background noise and low microphone volume.

Example

from transformers import pipeline

asr = pipeline("automatic-speech-recognition",
               model="SaarAI/Amharic-ASR-v2",
               device=0)
print(asr("audio.wav")["text"])                                       # uses the LM
print(asr("call.wav", chunk_length_s=20, stride_length_s=4)["text"])  # long audio
Read moreDetails
Benchmarks: average WER of 27.15% across 9 test sets (improved from 28.19% in v1).
Requirements: about 0.61 billion parameters for the acoustic model plus 613 MB for the KenLM language model.
Limitations: outputs lowercase text without punctuation; numbers are written out in Amharic words.
Discuss Telegram 0 3

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 2.3 GB · ready files3.0 GB
  • BF16weights 1.1 GB1.7 GB
  • FP8 / INT8recommendedweights 0.6 GB1.1 GB
  • 4-bitweights 0.3 GB0.8 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 · F32 — from RTX 2060

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

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.

    # SaarAI/Amharic-ASR-v2
    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 SaarAI/Amharic-ASR-v2 --local-dir ~/models/Amharic-ASR-v2
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Amharic-ASR-v2
  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

  1. w2v-bert-2.0family root
  2. Ethio-ASR-amharic
  3. Amharic-ASR-v1
  4. Amharic-ASR-v2

New posts about versions of w2v-bert-2.0 (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 1All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 1All on Hugging Face

About the model

SaarAI/Amharic-ASR-v2
Author
SaarAI
Category
Audio
Released
Posted

On NeyroLearn

Hugging Face

30

Ask AI about this model

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

Comments

Audio

Haru-4B

Haru-4B