SilencioNetwork/french-accents-speech

Accents of French — Silencio

French from 20 speakers born in 9 countries. 32 clips, 0.5 hours, each labelled with the speaker's country of birth, first language, self-reported accent or regional variety, declared French proficiency, age band, gender and recording device. Every clip ships with the prompt text the speaker read aloud. Human-validated transcription is available on request.

Hours	0.49
Clips	32
Speakers	20
Countries of birth	9
First languages	6
Self-reported accent / variety labels	10
Native French speakers (declared)	3 of 20
Audio	48 kHz mono and stereo WAV
Mean clip length	55.6 s
Prompt text	Included on every clip (prompt_text)
Transcripts	Human-validated transcription available on request
Licence	cc-by-nc-4.0

Most French speakers today live in Africa, and Silencio's French catalogue reflects that: 74% of hours in the September 2026 catalogue snapshot come from speakers born in Africa, led by Benin, Senegal, Cameroon, Morocco and Madagascar. France, Belgium, Switzerland and Canada together are 21%. This sample covers 9 countries of birth, led by Senegal and Tunisia. 17 of the 20 speakers use French as a second or additional language, most with Wolof or Arabic as their first language: the West African and Maghrebi French that most public French datasets underrepresent.

Each clip is a speaker reading a first-person monologue of about a minute, in their own voice and accent, on their own device. The text they were given is in prompt_text.

This is a sample, not the catalogue. It is drawn to show format, audio quality and metadata, and its speaker mix does not represent what Silencio holds. The off-the-shelf catalogue behind it is far more diverse across regions, age groups, accents and recording conditions. Men are 45% of the speakers in this sample. Across the catalogue the gender split is roughly 40% female to 60% male, with men in the majority in most languages. Licensed subsets can be drawn to a specified gender, age or regional balance.

Load it

from datasets import load_dataset

ds = load_dataset("SilencioNetwork/french-accents-speech", split="test")
print(ds[0]["country"], ds[0]["mother_tongue"], ds[0]["dialect"], ds[0]["proficiency"])

# datasets v4 returns a torchcodec AudioDecoder:
s = ds[0]["audio"].get_all_samples()
audio, sr = s.data, s.sample_rate

Requires pip install "datasets>=4.0" and FFmpeg ≥ 4.

Accents in this sample

By country of birth

Country of birth	Speakers	%
Senegal	8	40.0%
Tunisia	5	25.0%
Luxembourg	1	5.0%
Madagascar	1	5.0%
Niger	1	5.0%
Belgium	1	5.0%
South Africa	1	5.0%
Morocco	1	5.0%
Germany	1	5.0%

By first language

First language	Speakers	%
Wolof	8	40.0%
Arabic	6	30.0%
French	3	15.0%
Malagasy	1	5.0%
Hausa	1	5.0%
German	1	5.0%

Declared French proficiency

Proficiency	Speakers	%
fluent	14	70.0%
native	3	15.0%
conversational	2	10.0%
basic	1	5.0%

The self-reported accent or regional variety of every speaker is in the dialect column (10 distinct labels). It describes the speaker's own background, as they reported it at enrolment, and is left as recorded rather than corrected.

Speaker and recording metadata

Demographics

Gender	Speakers	%
female	11	55.0%
male	9	45.0%

Age band	Speakers	%
25-34	13	65.0%
35-44	3	15.0%
18-24	3	15.0%
45-59	1	5.0%

Recording conditions

Device	Clips	%
Mobile	21	65.6%
Desktop	11	34.4%

Splits

Single split, test, 32 rows. No train/dev/test partition is provided.

Speakers can contribute more than one clip. 8 speakers have 1 clip, 12 speakers have 2 clips. Any split you build must be speaker-disjoint, which means grouping on speaker_id rather than splitting by row, or the same voice will appear on both sides.

Fields

Column	Description	Values in this release
audio	Audio payload, stored at source rate and channel count	48 kHz mono and stereo WAV
speaker_id	Pseudonymous speaker identifier. Coherent within this dataset; deliberately not linkable to other Silencio releases	20 distinct
language	Language of the recording	constant: French
transcript	Empty in this release	Human-validated transcription available on request
transcript_type	Provenance of the transcript	constant: none
gender	Self-reported	female, male
country	Speaker's country of birth	9 distinct
mother_tongue	Speaker's self-reported first language	6 distinct
dialect	Self-reported accent / regional variety. The speaker's own background, NOT a classification of the recorded speech	10 distinct
os	Operating system of the recording device	Linux, Windows, macOS
device	Recording device class	Desktop, Mobile
duration	Seconds	32 distinct
script_type	Elicitation style	constant: read_monologue
age_band	Self-reported age, banded	18-24, 25-34, 35-44, 45-59
native_speaker	True where the speaker declares French as first language or at native level	2 distinct
proficiency	Speaker's self-declared proficiency in French	basic, conversational, fluent, native
prompt_text	Text the speaker was given to read aloud. Not a validated transcript: see Limitations	32 clips

What this is for

• Accent and first-language identification. Country of birth, first language and self-reported variety on every clip.
• Robustness and fairness auditing. Break ASR or speaker-model performance down by accent, first language, proficiency, age, gender and device.
• Speaker identification and verification across a wide spread of accents.
• Read-speech ASR checks on accented French, using prompt_text as approximate reference text. For a scored benchmark, human-validated transcription is available on request over this sample or any larger subset.
• Pronunciation and reading-fluency work. The same kind of prepared text read by speakers of many accents and proficiency levels.

Limitations

• Read speech. Speakers read prepared first-person monologues aloud. That keeps content comparable across accents, but it is not spontaneous conversation. Spontaneous French is available from the same contributor network.
• Uneven country counts. Some countries are represented by a single speaker; treat per-country figures in this sample as illustrative.
• Self-reported labels. Country, first language, accent and proficiency are declared by contributors, not assessed.
• Mixed channel layout. Audio is shipped as recorded. 8 clips are stereo with one silent channel; downmix to mono before analysis.
• Prompt text is not a transcript. prompt_text is the text the speaker was given, not a check of what they said. Speakers sometimes skip, repeat or rephrase. In 1 of 32 clips the reading rate falls outside 90 to 200 words per minute, which points to a partial or loose reading. Human-validated transcription of what was actually said is available on request.

French off the shelf

This release is a 0.5-hour sample. The accented French catalogue behind it holds more than 10,000 hours off the shelf, from contributors born in around 90 countries. Silencio publishes catalogue depth in bands rather than exact hour counts. Bands reflect the September 2026 off-the-shelf catalogue and move as collection continues.

Distribution by country of origin

Shares of hours and of contributors by country of origin, from a September 2026 catalogue snapshot:

Country of origin	Share of hours	Share of contributors
France	16.0%	11.5%
Benin	12.7%	13.6%
Senegal	10.4%	8.3%
Cameroon	8.6%	10.2%
Morocco	7.9%	5.5%
Madagascar	7.9%	8.0%
Togo	4.1%	6.9%
Algeria	4.1%	2.3%
Switzerland	3.5%	0.3%
Nigeria	3.2%	6.7%
Burkina Faso	2.9%	3.1%
Tunisia	2.9%	2.0%
Congo	2.3%	2.5%
Kenya	1.9%	4.1%
Andorra	1.8%	0.2%
75 further countries	9.7%	14.6%

Country of origin is self-reported and unverified; the smallest shares may include occasional enrolment or device-locale errors. Available by accent, country of origin, first language, proficiency, demographic profile and recording condition, as read or spontaneous speech. Human-validated transcription is available over any subset, scoped to your needs.

Provenance and consent

Every recording is contributed by an opted-in participant through the Silencio platform, under a consent record covering AI/ML training use. Contributors can request deletion, and deletion propagates to downstream releases. Full provenance documentation is available to licensees.

Speaker and recording identifiers in this release are pseudonymised afresh, so they do not match Silencio's internal IDs or those in any other Silencio release.

License

cc-by-nc-4.0: free for research and non-commercial use with attribution.

Attribution string: Silencio Network, Accents of French, 2026. CC BY-NC 4.0.

Non-commercial covers research, evaluation and publication. Benchmarking a commercial product model against this data is a commercial use and needs a licence. Ask; it is usually granted for evaluation. Model weights trained on this sample inherit the non-commercial restriction.

Commercial licensing, including terms for models trained on this data: silencio.network/contact

Citation

@misc{silencio_accents_french_2026,
  title  = {Accents of French — Silencio},
  author = {Silencio Network},
  year   = {2026},
  url    = {https://huggingface.co/datasets/SilencioNetwork/french-accents-speech}
}

Silencio: the catalogue behind this sample

This listing is a small fraction of what Silencio provides. Silencio specialises in underrepresented languages, accents and niche domains, delivered at scale. It draws on the largest global community of speech contributors and holds the largest off-the-shelf catalogue of niche-language speech available.

Recorded and available off the shelf. Audio already collected, with metadata, licensable today. Catalogue position as of September 2026:

Hours, off the shelf	around 500,000
Countries of contributor origin	around 180
Catalogue refresh	Weekly at record level, quarterly at catalogue level

Contributor community available on demand. Registered, consented contributors who can be activated for a specific brief:

Contributors available on demand	around 2,500,000
Countries	180+
Languages that can be collected	around 350

Anything not off the shelf can be sourced through the community. A language, an accent or dialect region, a demographic, a recording condition, a speech style or a specialist domain can be collected to a client's brief. Deep coverage across Africa, South-East Asia, South Asia and the Middle East.

Human-validated transcription is available for every language, scoped to each client's needs: script and orthography conventions, normalisation rules, word- or segment-level alignment, speaker labelling and turnaround. Every delivered clip is checked by a native-speaker reviewer. Published samples with human-validated transcripts: Kenyan Swahili, Cebuano, Tagalog / Filipino, Yoruba, Hausa and Amharic.

Proprietary and first-party. Every recording is collected directly by Silencio from consenting contributors. Nothing is scraped, and these recordings are not available in any other dataset on the internet.

Ethical sourcing, with provenance records. Every recording carries a consent record covering AI/ML training use, and contributor-level provenance documentation is available to licensees, including for EU AI Act training-data summaries. Contributors can withdraw consent, and withdrawal propagates to subsequent releases.

More at silencio.network.

For volume licensing, bespoke transcription or commissioned collection: silencio.network/contact