hviske-v5

A newer model is available — please use syvai/hviske-v5.3 instead. v5.3 is the current recommended Danish ASR model from this family and reaches 13.91% strict WER on the CoRal v3 full test set (beam=5). This v5 checkpoint is kept for reproducibility only.

Hviske v5 — Danish ASR

A 2B-parameter Conformer encoder-decoder ASR model trained for Danish speech recognition.

Trained on ~3.5M samples (~16,000 hours) of Danish speech across 7 datasets.

Training progression

Evaluation results

Evaluated on 200 samples per dataset (WER / CER):

Training details

• Architecture: 2B-param Conformer encoder-decoder, full fine-tuning (encoder + decoder), encoder BatchNorm in eval mode
• Optimizer: AdamW 8-bit, LR 2.5e-5, cosine schedule with 500-step warmup
• Batch size: 32 (4 × 8 gradient accumulation)
• Data mixing: Shuffle buffer (20k samples) across all datasets for balanced domain exposure
• Audio: 16kHz mono, max 20s per sample
• Hardware: Single NVIDIA RTX 3090 (24GB)

Training data

A unified version of all training data is available at syvai/danish-asr-unified.

Usage

Installation

pip install transformers==4.57.6 torch soundfile librosa

Basic transcription

from transformers import AutoProcessor, AutoModelForSpeechSeq2Seq
import soundfile as sf

processor = AutoProcessor.from_pretrained("syvai/hviske-v5", trust_remote_code=True)
model = AutoModelForSpeechSeq2Seq.from_pretrained("syvai/hviske-v5", trust_remote_code=True)
model = model.to("cuda")  # optional, for GPU inference

audio, sr = sf.read("audio.wav")

transcriptions = model.transcribe(
    processor=processor,
    audio_arrays=[audio],
    sample_rates=[sr],
    language="da",
    punctuation=True,
)
print(transcriptions[0])

Batch transcription

import soundfile as sf

files = ["file1.wav", "file2.wav", "file3.wav"]
arrays, rates = [], []
for f in files:
    audio, sr = sf.read(f)
    arrays.append(audio)
    rates.append(sr)

transcriptions = model.transcribe(
    processor=processor,
    audio_arrays=arrays,
    sample_rates=rates,
    language="da",
    punctuation=True,
)
for f, t in zip(files, transcriptions):
    print(f"{f}: {t}")

High-throughput inference with vLLM

For production workloads, serve the model with vLLM for significantly higher throughput:

pip install vllm
vllm serve syvai/hviske-v5 --trust-remote-code

Then send requests:

import requests, base64, soundfile as sf, io

audio, sr = sf.read("audio.wav")
buf = io.BytesIO()
sf.write(buf, audio, sr, format="WAV")
audio_b64 = base64.b64encode(buf.getvalue()).decode()

response = requests.post("http://localhost:8000/v1/chat/completions", json={
    "model": "syvai/hviske-v5",
    "messages": [{"role": "user", "content": [
        {"type": "input_audio", "input_audio": {"data": audio_b64, "format": "wav"}}
    ]}],
})
print(response.json()["choices"][0]["message"]["content"])

License

This model is released under Creative Commons Attribution-NonCommercial 4.0 (CC BY-NC 4.0).

• Permitted: non-commercial use including research, education, evaluation, and personal projects, with attribution.
• Not permitted without a separate commercial license: any use by or for a commercial entity, integration into a commercial product or service, or use to generate revenue (directly or indirectly).
• Commercial licensing: contact [email protected].