GigaAM-v3: transcribe.cpp GGUF

GGUF conversions of ai-sage/GigaAM-v3 for use with transcribe.cpp.

Ported from upstream commit ec1dc1f, pinned 2026-05-12. Validated against the gigaam author package reference at transcribe.cpp commit 0fff13f8 on 2026-10-15.

Offline Russian speech-to-text with greedy RNN-T decoding. 16-layer Conformer encoder paired with an RNN-T transducer head. Output is cased Russian with punctuation, decoded from a 1024-piece SentencePiece tokenizer. Not a streaming model and does not translate. Audio past the 25 s training window is encoded in overlapping windows.

Downloads

WER on the full FLEURS ru split (775 utterances), batch sizes 1 and 8, timestamps none. Figures without a commit were published before provenance was recorded.

Greedy decoding, no external LM. F32 reference baseline: 5.35%. Upstream gigaam author package measured on the same manifest: 6.78%; the 1.4 pp gap is upstream rejecting 5 long (>25 s) utterances with Too long wav file, use 'transcribe_longform' method. (counted as 100% deletion errors). On the 770-utt subset both sides decode, transcribe.cpp matches upstream exactly. ai-sage does not publish a FLEURS ru WER; this number is measured here.

Usage

Build transcribe.cpp from source:

git clone git@github.com:handy-computer/transcribe.cpp.git
cd transcribe.cpp
cmake -B build && cmake --build build

Run on a 16 kHz mono WAV:

build/bin/transcribe-cli \
  -m gigaam-v3-e2e-rnnt-Q8_0.gguf \
  input.wav

If your audio isn't already 16 kHz mono WAV, convert it first:

ffmpeg -i input.mp3 -ar 16000 -ac 1 output.wav

See the transcribe.cpp model page for performance numbers, numerical validation, and reproduction steps.

License

Inherited from the base model: MIT. See the upstream model card for full terms.

Original Model Card

The section below is reproduced from ai-sage/GigaAM-v3 at commit ec1dc1f for offline reference. The upstream card is the authoritative source.

GigaAM-v3

GigaAM-v3 is a Conformer-based foundation model with 220–240M parameters, pretrained on diverse Russian speech data using the HuBERT-CTC objective. It is the third generation of the GigaAM family and provides state-of-the-art performance on Russian ASR across a wide range of domains.

GigaAM-v3 includes the following model variants:

• ssl — self-supervised HuBERT–CTC encoder pre-trained on 700,000 hours of Russian speech
• ctc — ASR model fine-tuned with a CTC decoder
• rnnt — ASR model fine-tuned with an RNN-T decoder
• e2e_ctc — end-to-end CTC model with punctuation and text normalization
• e2e_rnnt — end-to-end RNN-T model with punctuation and text normalization

GigaAM-v3 training incorporates new internal datasets: callcenter conversations, speech with background music, natural speech, and speech with atypical characteristics. the models perform on average 30% better on these new domains, while maintaining the same quality as previous GigaAM generations on public benchmarks.

The table below reports the Word Error Rate (%) for GigaAM-v3 and other existing models over diverse domains.

The end-to-end ASR models (e2e_ctc and e2e_rnnt) produce punctuated, normalized text directly. In end-to-end ASR comparisons of e2e_ctc and e2e_rnnt against Whisper-large-v3, using Gemini 2.5 Pro as an LLM-as-a-judge, GigaAM-v3 models win by an average margin of 70:30.

For detailed results, see metrics.

Usage

from transformers import AutoModel

revision = "e2e_rnnt"  # can be any v3 model: ssl, ctc, rnnt, e2e_ctc, e2e_rnnt
model = AutoModel.from_pretrained(
    "ai-sage/GigaAM-v3",
    revision=revision,
    trust_remote_code=True,
)

transcription = model.transcribe("example.wav")
print(transcription)

Recommended versions:

• torch==2.8.0, torchaudio==2.8.0
• transformers==4.57.1
• pyannote-audio==4.0.0, torchcodec==0.7.0
• (any) hydra-core, omegaconf, sentencepiece

Full usage guide can be found in the example.

License: MIT

Paper: GigaAM: Efficient Self-Supervised Learner for Speech Recognition (InterSpeech 2025)