README.md

---
language:
- en

license: unknown

task_categories:
- automatic-speech-recognition
- audio-classification
- text-classification

annotations_creators:
- expert-generated

source_datasets:
- earnings22

multilinguality:
- monolingual

size_categories:
- 1K<n<10K

pretty_name: "HALAS: A Human-Annotated Dataset of Hallucinations of Modern ASR Systems"

tags:
- asr
- speech-recognition
- speech-processing
- hallucination
- hallucination-detection
- whisper
- earnings22
- benchmark
- human-annotations
- error-detection
- audio
- speech-to-text
- robustness

dataset_info:
  - config_name: default
    features:
  - name: audio_id
    dtype: string

  - name: audio_duration
    dtype: float32

  - name: e22_reference_text
    dtype: string

  - name: corrected_reference_text
    dtype: string

  - name: whisper_large_v2_prediction
    dtype: string
  - name: whisper_large_v2_label
    dtype: bool
  - name: whisper_large_v2_hallucination_text
    dtype: string
  - name: whisper_large_v2_hallucination_json
    dtype: string

  - name: whisper_large_v3_prediction
    dtype: string
  - name: whisper_large_v3_label
    dtype: bool
  - name: whisper_large_v3_hallucination_text
    dtype: string
  - name: whisper_large_v3_hallucination_json
    dtype: string

  - name: whisper_large_v3_turbo_prediction
    dtype: string
  - name: whisper_large_v3_turbo_label
    dtype: bool
  - name: whisper_large_v3_turbo_hallucination_text
    dtype: string
  - name: whisper_large_v3_turbo_hallucination_json
    dtype: string

  - name: crisper_whisper_prediction
    dtype: string
  - name: crisper_whisper_label
    dtype: bool
  - name: crisper_whisper_hallucination_text
    dtype: string
  - name: crisper_whisper_hallucination_json
    dtype: string

  - name: canary_prediction
    dtype: string
  - name: canary_label
    dtype: bool
  - name: canary_hallucination_text
    dtype: string
  - name: canary_hallucination_json
    dtype: string

  - name: canary_flash_prediction
    dtype: string
  - name: canary_flash_label
    dtype: bool
  - name: canary_flash_hallucination_text
    dtype: string
  - name: canary_flash_hallucination_json
    dtype: string

  - name: parakeet_prediction
    dtype: string
  - name: parakeet_label
    dtype: bool
  - name: parakeet_hallucination_text
    dtype: string
  - name: parakeet_hallucination_json
    dtype: string

  - name: phi4_prediction
    dtype: string
  - name: phi4_label
    dtype: bool
  - name: phi4_hallucination_text
    dtype: string
  - name: phi4_hallucination_json
    dtype: string

  - name: granite_prediction
    dtype: string
  - name: granite_label
    dtype: bool
  - name: granite_hallucination_text
    dtype: string
  - name: granite_hallucination_json
    dtype: string

  - name: split
    dtype: string

    splits:
      - name: train
        num_examples: 2866

      - name: test
        num_examples: 745

configs:
  - config_name: default
    data_files:
      - split: train
        path: train.csv
      - split: test
        path: test.csv

paperswithcode_id: null

train_eval_index:
- config: default
  task: hallucination-detection
  split: test

---
# Dataset Card for HALAS

## Dataset Summary

**HALAS (Hallucination Annotations for Large-scale ASR Systems)** is a human-annotated dataset of hallucinations produced by modern automatic speech recognition (ASR) systems on real-world speech recordings. The dataset contains span-level hallucination annotations for ASR outputs generated from recordings in the Earnings22 corpus.

HALAS was introduced to address a key limitation in prior hallucination research: most existing hallucination detection and mitigation methods are evaluated on non-speech audio or artificially corrupted speech. In contrast, HALAS captures **naturally occurring hallucinations on authentic speech recordings**, enabling realistic evaluation of hallucination detection, mitigation, and analysis methods.

The dataset contains annotations of ASR outputs generated by multiple state-of-the-art speech recognition systems and provides both utterance-level labels and span-level annotations identifying hallucinated content.

HALAS is intended as a benchmark for:

- ASR hallucination detection
- Hallucination mitigation research
- Error analysis of speech recognition systems
- Evaluation of reference-based and reference-free hallucination metrics
- Development of robust ASR systems

---

## Dataset Details

The dataset contains **3,611 audio files** from the Earnings22 dataset with manually verified and corrected annotations.

For each audio segment, predictions from multiple state-of-the-art ASR models were collected and independently reviewed by human annotators. Hallucinated spans, looping errors, and hallucination-looping errors were labeled at the text-span level.

In addition, reference transcripts provided by Earnings22 were manually revalidated and corrected when necessary.

---

## Dataset Description

HALAS consists of human-annotated hallucinations occurring in ASR transcriptions generated from recordings in the Earnings22 dataset.

https://huggingface.co/datasets/distil-whisper/earnings22



The dataset was created using outputs from state-of-the-art ASR systems applied to naturally occurring earnings-call recordings containing speakers from 27 countries and a wide range of recording conditions.

To maximize the prevalence of hallucinations in the annotation pool, candidate audio segments were selected using **inter-model disagreement**. Predictions from multiple ASR systems were compared using average pairwise Word Error Rate (WER), and segments with the highest disagreement were prioritized for annotation.

The annotation process focused exclusively on the relationship between the audio signal and the ASR prediction. Annotators were instructed **not** to rely on the reference transcript when identifying hallucinations.

A hallucination was defined as:

> A prediction, or fragment thereof, that has no phonetic correspondence with the content of the audio signal analyzed by the ASR model.

Annotators marked words or phrases using the following taxonomy:

### Hallucination

A word or phrase that does not correspond to any spoken content in the audio recording.

### Looping

An erroneous repetition of a word or phrase that is present in the speech signal.

### Hallucination Looping

An erroneous repetition of content that is itself hallucinated and does not correspond to the audio signal.

Each audio file was annotated independently by two annotators, with disagreements resolved by a third annotator acting as an arbitrator.

In addition to hallucination annotation, all reference transcripts were manually reviewed and corrected when necessary. During this process, approximately 14% of examples were identified as partially inaudible or ambiguous.

The resulting dataset contains:

- Human-annotated hallucination spans
- Human-annotated looping spans
- Human-annotated hallucination-looping spans
- Corrected reference transcripts
- Utterance-level hallucination labels
- Predictions from multiple ASR systems

Importantly, HALAS was intentionally constructed to contain a high proportion of hallucinations and therefore **does not represent the real-world prevalence of ASR hallucinations**. However, all hallucinations originate from genuine speech recordings rather than synthetic corruption or non-speech inputs.

---

## Supported Tasks

- Automatic Speech Recognition (ASR)
- Hallucination Detection
- Hallucination Mitigation
- Error Detection
- Speech Recognition Evaluation
- Speech Processing Research

---

## Languages

- English

The source recordings originate from earnings-call recordings featuring speakers from 27 countries.

---

## License

HALAS is derived from the Earnings22 dataset and augments it with human-generated hallucination annotations and corrected reference transcripts.

The original Earnings22 dataset is publicly distributed under the **Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0)** license. Users of HALAS must comply with the terms of the Earnings22 license in addition to any conditions specified for this dataset. The full license text is available at:

https://creativecommons.org/licenses/by-sa/4.0/

HALAS contains original human annotations produced by the authors. Unless otherwise specified in the official repository, these annotations are intended to be distributed under the same license as the source dataset for compatibility with the underlying Earnings22 data.

**Note:** Users should consult the official HALAS repository for the authoritative licensing terms of the released version.

---

## Dataset Sources

### Repository

https://github.com/DSP-AGH/HALAS/tree/main

### Paper

**HALAS: A Human-Annotated Dataset of Hallucinations of Modern ASR Systems**

Mateusz Barański, Jan Jasiński, Julitta Bartolewska, Marcin Witkowski, and Konrad Kowalczyk.

Accepted at Interspeech 2026.

---

## Data Fields

The dataset contains the following fields:

| Field | Description |
|---------|-------------|
| `audio_id` | Unique identifier of the audio segment from Earnings22 |
| `audio_duration` | Duration of the audio segment in seconds |
| `e22_reference_text` | Original reference transcript provided by Earnings22 |
| `corrected_reference_text` | Human-corrected reference transcript |
| `model_prediction` | ASR prediction generated by the corresponding model |
| `model_label` | Utterance-level hallucination label indicating whether hallucination was present |
| `model_hallucination_text` | Text span annotated as hallucination |
| `model_hallucination_json` | Structured span annotations containing hallucination locations and categories |
| `split` | Dataset split (`train` or `test`) |

The `model` field corresponds to one of the following ASR systems:

- `whisper_large_v2`
- `whisper_large_v3`
- `whisper_large_v3_turbo`
- `crisper_whisper`
- `canary`
- `canary_flash`
- `parakeet`
- `phi4`*
- `granite`*


Models marked with an asterisk (*) are outside the scope of related paper, as they were trained on the Earnings22 dataset.

---

## Annotation Methodology

### Annotators

HALAS was annotated by **10 professional annotators** with English proficiency at B2 level or higher.

### Annotation Procedure

For each audio sample:

1. The audio recording was reviewed.
2. Audio quality issues were identified (e.g., non-English speech, overlapping speakers, inaudible segments).
3. Annotators reviewed ASR predictions while listening to the audio.
4. Hallucinated spans were marked using the HALAS taxonomy.
5. A second annotator independently repeated the process.
6. A third annotator resolved disagreements and finalized annotations.

### Quality Control

Each sample was reviewed by two independent annotators.

Agreement between annotators was high:

- Cohen's κ = 0.87

This level of agreement indicates strong consistency in hallucination identification.

Additionally, all reference transcripts were revalidated and corrected by the arbitrator.

---

## Data Splits

| Split | Files |
|---------|---------|
| Train | 2,866 |
| Test | 745 |

The train/test partition was created by:

- Splitting by source meeting
- Stratifying according to:
  - Average WER
  - Hallucination rate
  - Audio duration

To ensure a stable benchmark, the test set includes only recordings:

- Longer than 1 second
- Containing at least 3 words

Dataset statistics:

| Split | Hallucination Rate |
|---------|---------|
| Train | 33.6% |
| Test | 22.6% |

---

## Dataset Characteristics

### Source Audio

The source recordings originate from the Earnings22 dataset:

- Approximately 119 hours of audio
- Earnings-call recordings
- Speakers from 27 countries
- Real-world recording conditions

Candidate segments were selected from portions exhibiting high disagreement between ASR systems.

Audio files can be downloaded using following script:

```
import csv
from datasets import load_dataset
import soundfile as sf
import os 

# Load dataset
filtered_dataset = load_dataset("distil-whisper/earnings22", 'chunked', split='test')

# Create output directory
output_dir = "e22"
os.makedirs(output_dir, exist_ok=True)

# CSV file setup
csv_file = os.path.join(output_dir, "metadata.csv")

with open(csv_file, mode='w', newline='', encoding='utf-8') as file:
    writer = csv.writer(file)
    writer.writerow(["filepath", "text"])  # Write header
    
    for audio in filtered_dataset:
        seg_id = audio["segment_id"]
        audio_array = audio["audio"]["array"]
        sample_rate = audio["audio"]["sampling_rate"]
        name = audio["file_id"]
        name = "_".join([seg_id, name])
        file_path = os.path.join(output_dir, name + ".wav")
        text = audio['transcription']
        
        # Save audio file
        sf.write(file_path, audio_array, sample_rate)
        
        # Write metadata to CSV
        writer.writerow([file_path, text])
        print(f"Saved: {file_path}")

print("All files and metadata saved successfully.")

```

---

## Bias, Risks, and Limitations

HALAS intentionally oversamples difficult audio segments.

As a consequence:

1. The hallucination frequency does not reflect real-world deployment conditions.
2. The dataset is biased toward audio examples that produce disagreement among ASR systems.
3. Results obtained on HALAS should not be interpreted as estimates of hallucination prevalence in production systems.
4. The dataset focuses exclusively on English earnings-call recordings and may not generalize to other domains, languages, or acoustic environments.
5. Some audio samples contain challenging conditions such as accents, low-quality recordings, or partially inaudible speech.
6. Although references were manually corrected, certain recordings remain inherently ambiguous due to poor audio quality.

Researchers should therefore view HALAS primarily as a benchmark for hallucination detection and analysis rather than a representative sample of everyday ASR usage.

---

## Recommendations

Researchers using HALAS should:

1. Report whether evaluations are performed at utterance level or span level.
2. Account for the intentionally elevated hallucination rate when interpreting results.
3. Distinguish hallucinations from ordinary transcription errors.
4. Consider both structural and semantic evaluation metrics.
5. Evaluate generalization on additional datasets whenever possible.

Because HALAS contains naturally occurring hallucinations from real speech recordings, it provides a substantially more realistic benchmark than datasets constructed from non-speech or artificially corrupted audio.

---

## Citation

### BibTeX

```bibtex
@inproceedings{baranski2026halas,
  title={HALAS: A Human-Annotated Dataset of Hallucinations of Modern ASR Systems},
  author={Barański, Mateusz and Jasiński, Jan and Bartolewska, Julitta and Witkowski, Marcin and Kowalczyk, Konrad},
  booktitle={Proceedings of Interspeech 2026},
  year={2026}
}
```

### APA

Barański, M., Jasiński, J., Bartolewska, J., Witkowski, M., & Kowalczyk, K. (2026). *HALAS: A Human-Annotated Dataset of Hallucinations of Modern ASR Systems*. In *Proceedings of Interspeech 2026*.

---

## Dataset Card Authors

Mateusz Barański, Jan Jasiński, Julitta Bartolewska, Marcin Witkowski, and Konrad Kowalczyk.

Signal Processing Group  
Institute of Electronics  
AGH University of Krakow, Poland

---

## Dataset Card Contact

- Mateusz Barański — mbaranski@agh.edu.pl
- Jan Jasiński — jjasinsk@agh.edu.pl

Alternatively, please use the issue tracker in the official repository once available.