Whisper-Small Dutch - Mixed Synthetic Data (Mid-High Quality Filtered)

This model is a fine-tuned version of openai/whisper-small for Dutch automatic speech recognition (ASR). It was trained on Common Voice 17.0 Dutch combined with WAVe-filtered synthetic speech data (quality threshold q ≥ 0.5).

Introduction

How the Data Was Created

The training data combines real speech from Common Voice 17.0 with synthetic speech generated through a two-stage pipeline:

Transcript Generation: We used GPT-4o-mini to generate Dutch transcripts that match the word count distribution observed in Common Voice, ensuring realistic utterance lengths and diverse linguistic content.
Speech Synthesis: Each transcript was converted to audio using OpenAI's TTS-1 model with 9 different voice variants (alloy, ash, coral, echo, fable, nova, onyx, sage, shimmer), producing 34,898 synthetic samples.
Quality Filtering with WAVe: Raw synthetic speech often contains defects such as mispronunciations, omitted words, or prosodic anomalies. To address this, we applied WAVe (Word-Aligned Verification), a model that assesses audio-text alignment at the word level rather than the sentence level. WAVe uses multi-head attention to align each word to its corresponding audio frames and assigns per-word confidence scores via a GLU-based scorer. Samples scoring below the threshold (q < 0.5) were removed, retaining 30,182 high-quality synthetic samples.

How the Model Was Created

The model was fine-tuned from openai/whisper-small using the Hugging Face Transformers library with the following approach:

Mixed Training: Combined 34,952 real speech samples from Common Voice 17.0 Dutch with 30,182 WAVe-filtered synthetic samples (65,134 total).
Optimization: Trained for 5 epochs with a learning rate of 1e-5, global batch size of 256, and BF16 precision on an NVIDIA H200 GPU.
Checkpoint Selection: The best checkpoint was selected based on validation loss, occurring at step 500 with a validation loss of 0.1484.

This approach achieves the best Test WER (10.86%) among all Whisper-Small Dutch configurations while maintaining strong cross-domain generalization.

Model Details

Property	Value
Base Model	openai/whisper-small
Language	Dutch (nl)
Task	Automatic Speech Recognition (transcribe)
Parameters	244M
Training Data	Common Voice 17.0 + Mid-High Quality Synthetic (q ≥ 0.5)
Total Training Samples	65,134
Sampling Rate	16kHz

Evaluation Results

This Model (whisper-small-mixed-cv-nl)

Metric	Value
Validation Loss	0.1484
Validation WER	8.73%
Test WER (Common Voice)	10.86%
Test WER (MLS)	30.04%
Best Checkpoint	Step 500
Max Training Steps	1,270

Comparison with Other Training Configurations (Whisper-Small Dutch)

Training Data	Max Steps	Val Loss	Val WER	Test WER (CV)	Test WER (MLS)
Common Voice Only	680	0.1491	8.73%	11.13%	30.71%
High-Quality Filtered + CV	890	0.1493	8.76%	11.00%	29.91%
Mid-High Quality Filtered + CV	1,270	0.1484	8.73%	10.86%	30.04%
All Synthetic + CV (Unfiltered)	1,365	0.1484	8.64%	10.91%	30.06%

Key Performance Highlights

Best Test WER (10.86%) on Common Voice among all Whisper-Small Dutch configurations
2.4% relative improvement on Common Voice test set vs baseline (10.86% vs 11.13%)
2.2% relative improvement on MLS benchmark vs baseline (30.04% vs 30.71%)
7% fewer training steps than unfiltered synthetic data while achieving better in-domain performance

Training Data

Dataset Composition

Source	Samples	Description
Common Voice 17.0 Dutch	34,952	Real speech from Mozilla's crowdsourced dataset
Synthetic Transcript NL (q ≥ 0.5)	30,182	WAVe-filtered TTS audio from GPT-4o-mini transcripts
Total	65,134

Synthetic Data Generation Pipeline

The synthetic dataset (yuriyvnv/synthetic_transcript_nl) was generated using:

Transcript Generation: GPT-4o-mini, matching Common Voice word count distribution
Speech Synthesis: OpenAI TTS-1 model with 9 voice variants (alloy, ash, coral, echo, fable, nova, onyx, sage, shimmer)
Quality Filtering: WAVe model filtering at threshold q ≥ 0.5

WAVe Quality Distribution (Dutch Synthetic Data)

Quality Level	Samples	Percentage	Used in This Model
High (q ≥ 0.8)	10,555	30.2%	✓
Medium (0.5 ≤ q < 0.8)	19,627	56.2%	✓
Low (q < 0.5)	4,716	13.5%	✗

Training Procedure

Hyperparameters

Parameter	Value
Learning Rate	1e-5
Batch Size (Global)	256
Warmup Steps	200
Max Epochs	5
Precision	BF16
Optimizer	AdamW (fused)
Eval Steps	50
Metric for Best Model	eval_loss

Training Infrastructure

GPU: NVIDIA H200 (140GB VRAM)
Operating System: Ubuntu 22.04
Framework: Hugging Face Transformers

Training Curve

Step  100: val_loss = 0.1946
Step  250: val_loss = 0.1625
Step  400: val_loss = 0.1544
Step  500: val_loss = 0.1484 ← Best checkpoint
Step  750: val_loss = 0.1484
Step 1000: val_loss = 0.1522
Step 1250: val_loss = 0.1552

Usage

Transcription Pipeline

from transformers import pipeline

transcriber = pipeline(
    "automatic-speech-recognition",
    model="yuriyvnv/whisper-small-mixed-cv-nl",
    device="cuda"
)

result = transcriber("path/to/dutch_audio.wav")
print(result["text"])

Direct Model Usage

from transformers import WhisperProcessor, WhisperForConditionalGeneration
import librosa

processor = WhisperProcessor.from_pretrained("yuriyvnv/whisper-small-mixed-cv-nl")
model = WhisperForConditionalGeneration.from_pretrained("yuriyvnv/whisper-small-mixed-cv-nl")
model.to("cuda")

audio, sr = librosa.load("path/to/dutch_audio.wav", sr=16000)
input_features = processor(audio, sampling_rate=16000, return_tensors="pt").input_features.to("cuda")

predicted_ids = model.generate(input_features)
transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]
print(transcription)

Specifying Language

model.generation_config.language = "nl"
model.generation_config.task = "transcribe"

Methodology

This model leverages WAVe (Word-Aligned Verification), a word-level quality assessment method for filtering synthetic speech data. Unlike sentence-level filtering approaches, WAVe:

Aligns each word to its corresponding audio frames using multi-head attention
Assigns per-word confidence scores via a GLU-based scorer
Detects localized synthesis errors (mispronunciations, omitted words, prosodic anomalies)
Achieves 6.5% improvement over sentence-level filtering methods

For full methodology details, see the references below.

When to Use This Model

This model is ideal when:

Best in-domain accuracy is required: Achieves 10.86% Test WER (best among Small Dutch models)
Balanced performance: Good tradeoff between in-domain and cross-domain generalization
Moderate compute budget: 7% fewer steps than unfiltered approach

Consider alternatives based on your needs:

whisper-small-high-mixed-nl: 35% fewer steps, slight accuracy tradeoff
whisper-small-cv-fully-synthetic-nl: Maximum data, similar performance

Limitations

Domain specificity: Optimized for general Dutch; may underperform on technical domains
Acoustic conditions: Trained on clean speech; noise robustness not guaranteed
Dialect coverage: Performance may vary across Dutch regional variants

Citation

@article{perezhohin2024enhancing,
  title={Enhancing Automatic Speech Recognition: Effects of Semantic Audio Filtering on Models Performance},
  author={Perezhohin, Yuriy and Santos, Tiago and Costa, Victor and Peres, Fernando and Castelli, Mauro},
  journal={IEEE Access},
  year={2024},
  publisher={IEEE}
}