VoxKnesset

Voice recordings of Israeli politicians from Knesset proceedings, annotated with speaker age and demographic metadata.

Dataset Summary

• Total hours (longitudinal subset): 2,307
• Plenary sessions: ~1,550
• Unique speakers: 393 Members of Knesset
• Language: Hebrew
• Recording years: 2009–2025 (16 years)
• Maximum span per speaker: 15 years
• Median span per speaker: 3.4 years
• Speakers with >10 years coverage: 47 (12%)
• Age range: 28–81 years

Split	Samples	Speakers
train	49,433	314
test	17,237	79

Audio: 16-bit PCM, mono, 16 kHz WAV files.

Dataset Curation

VoxKnesset was constructed from 16 years (2009–2025) of official Israeli parliamentary plenary recordings. Raw audiovisual files (~8,800 hours) were processed through a multi-stage pipeline: audio was extracted at 16 kHz, timestamps were corrected, and word-level forced alignment was performed against official human-verified transcripts. Speaker-attributed segments were then matched using structured parliamentary records, filtered for high alignment confidence and textual similarity, and restricted to segments of sufficient duration. The final longitudinal subset contains 2,307 hours of single-speaker Hebrew speech from 393 Members of Knesset, each linked to verified demographic metadata (birth year, gender, religion, and birthplace).

Features

Column	Type	Description
audio	Audio	Raw waveform (16 kHz)
speaker_id	int	Unique speaker identifier
age	float	Speaker age at time of recording (years)
gender	int	Gender (0 = female, 1 = male)
speaker_place_of_birth	string	Place of birth (e.g. ישראל, מרוקו, עיראק, ברית המועצות, פולין, ארצות הברית)
speaker_year_of_aliya	string	Year of immigration to Israel (e.g. 1948, 1956, 1969)
speaker_religion	string	Religion (יהודי, מוסלמי, נוצרי, דרוזי)
speaker_nationality	string	Nationality (יהודי, ערבי, דרוזי, בדואי)
speaker_religious_orientation	string	Religious orientation (חילוני, דתי, חרדי, סוני, קתולי, יווני-קתולי)

Transcripts

A separate transcripts.parquet file provides the official Knesset protocol text aligned to each audio segment (66,670 entries).

Column	Type	Description
filename	string	WAV filename (e.g. 433_2120_10152_10474.wav)
text	string	Hebrew transcript from the Knesset protocol

The filename encodes {speaker_id}_{plenum_id}_{start_seconds}_{end_seconds}.wav and matches the path field inside the audio column of the main dataset.

Usage

Load audio with transcripts (combined)

import os
import pandas as pd
from datasets import load_dataset

# Load audio + metadata
ds = load_dataset("yanirmr/VoxKnesset")

# Load transcript text
transcripts = pd.read_parquet(
    "hf://datasets/yanirmr/voxknesset/transcripts.parquet"
)
text_lookup = dict(zip(transcripts["filename"], transcripts["text"]))

# Combine: audio, metadata, and transcript
sample = ds["train"][0]
filename = os.path.basename(sample["audio"]["path"])

print(f"File:    {filename}")
print(f"Speaker: {sample['speaker_id']}  Age: {sample['age']:.1f}  Gender: {'M' if sample['gender'] == 1 else 'F'}")
print(f"Text:    {text_lookup[filename][:120]}...")
print(f"Audio:   {len(sample['audio']['array'])} samples @ {sample['audio']['sampling_rate']} Hz")

Streaming (no full download)

import os
import pandas as pd
from datasets import load_dataset

ds = load_dataset("yanirmr/VoxKnesset", split="train", streaming=True)

transcripts = pd.read_parquet(
    "hf://datasets/yanirmr/voxknesset/transcripts.parquet"
)
text_lookup = dict(zip(transcripts["filename"], transcripts["text"]))

for sample in ds.take(3):
    filename = os.path.basename(sample["audio"]["path"])
    text = text_lookup.get(filename, "")
    print(f"{filename}  speaker={sample['speaker_id']}  age={sample['age']:.1f}")
    print(f"  {text[:100]}...")

Load only the transcripts (no audio)

import pandas as pd

transcripts = pd.read_parquet(
    "hf://datasets/yanirmr/voxknesset/transcripts.parquet"
)
print(f"{len(transcripts):,} transcripts")
print(transcripts.head())

Citation

If you use our dataset, the following quote is preferable:

@misc{marmor2026voxknessetlargescalelongitudinalhebrew,
      title={VoxKnesset: A Large-Scale Longitudinal Hebrew Speech Dataset for Aging Speaker Modeling}, 
      author={Yanir Marmor and Arad Zulti and David Krongauz and Adam Gabet and Yoad Snapir and Yair Lifshitz and Eran Segal},
      year={2026},
      eprint={2603.01270},
      archivePrefix={arXiv},
      primaryClass={eess.AS},
      url={https://arxiv.org/abs/2603.01270}, 
}

License

The dataset is released under the ivrit.ai License, which enables broad research and commercial use.

• Full license: https://www.ivrit.ai/en/the-license/
• FAQs: https://www.ivrit.ai/en/license-faqs/