yuriyvnv/parakeet-tdt-0.6b-dutch

Parakeet-TDT-0.6B Dutch

A Dutch automatic speech recognition (ASR) model fine-tuned from nvidia/parakeet-tdt-0.6b-v3.

Model Details

Property	Value
Base model	nvidia/parakeet-tdt-0.6b-v3
Architecture	FastConformer-TDT (600M params)
Language	Dutch (nl)
Input	16 kHz mono audio
Output	Dutch text with punctuation and capitalization
License	CC-BY-4.0

Evaluation Results

Evaluated on Common Voice 17.0 Dutch splits (raw text, no normalization):

Split	WER	CER	Samples
Validation	3.73%	1.02%	9,062
Test	5.33%	1.46%	11,266

Training

Fine-tuned on a combination of:

• Common Voice 17.0 (nl) -- human-recorded Dutch speech
• Synthetic Transcript NL -- 34,898 synthetic Dutch speech samples generated with OpenAI TTS

Training Configuration

Parameter	Value
Optimizer	AdamW
Learning rate	5e-5 (cosine annealing)
Warmup	10% of total steps
Batch size	64
Precision	bf16-mixed
Gradient clipping	1.0
Early stopping	10 epochs patience on val WER
Best epoch	21

Usage

Installation

pip install nemo_toolkit[asr]

Transcribe Audio

import nemo.collections.asr as nemo_asr

# Load model
asr_model = nemo_asr.models.ASRModel.from_pretrained(
    model_name="yuriyvnv/parakeet-tdt-0.6b-dutch"
)

# Transcribe
output = asr_model.transcribe(["audio.wav"])
print(output[0].text)

Transcribe with Timestamps

output = asr_model.transcribe(["audio.wav"], timestamps=True)

for stamp in output[0].timestamp["segment"]:
    print(f"{stamp['start']:.1f}s - {stamp['end']:.1f}s : {stamp['segment']}")

Long-Form Audio

For audio longer than 24 minutes, enable local attention:

asr_model.change_attention_model(
    self_attention_model="rel_pos_local_attn",
    att_context_size=[256, 256],
)
output = asr_model.transcribe(["long_audio.wav"])

Intended Use

This model is designed for transcribing Dutch speech to text. It works best on:

• Read speech and conversational Dutch
• Audio recorded at 16 kHz or higher
• Segments up to 24 minutes (or longer with local attention enabled)

Limitations

• Trained primarily on European Portuguese-accented Dutch from Common Voice; performance may vary on regional dialects or heavily accented speech
• Synthetic training data was generated with OpenAI TTS voices, which may not fully represent natural speech variability
• Not suitable for real-time streaming without additional configuration