Dataset Viewer
Duplicate
The dataset viewer is not available for this split.
Cannot extract the features (columns) for the split 'train' of the config 'conversation' of the dataset.
Error code:   FeaturesError
Exception:    ArrowInvalid
Message:      JSON parse error: Invalid value. in row 0
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 342, in _generate_tables
                  df = pandas_read_json(f)
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 38, in pandas_read_json
                  return pd.read_json(path_or_buf, **kwargs)
                         ~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/pandas/io/json/_json.py", line 791, in read_json
                  json_reader = JsonReader(
                      path_or_buf,
                  ...<16 lines>...
                      engine=engine,
                  )
                File "/usr/local/lib/python3.14/site-packages/pandas/io/json/_json.py", line 905, in __init__
                  self.data = self._preprocess_data(data)
                              ~~~~~~~~~~~~~~~~~~~~~^^^^^^
                File "/usr/local/lib/python3.14/site-packages/pandas/io/json/_json.py", line 917, in _preprocess_data
                  data = data.read()
                File "/usr/local/lib/python3.14/site-packages/datasets/utils/file_utils.py", line 838, in read_with_retries
                  out = read(*args, **kwargs)
                File "<frozen codecs>", line 325, in decode
              UnicodeDecodeError: 'utf-8' codec can't decode byte 0x9a in position 14: invalid start byte
              
              During handling of the above exception, another exception occurred:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/split/first_rows.py", line 244, in compute_first_rows_from_streaming_response
                  iterable_dataset = iterable_dataset._resolve_features()
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 4645, in _resolve_features
                  features = _infer_features_from_batch(self.with_format(None)._head())
                                                        ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2866, in _head
                  return next(iter(self.iter(batch_size=n)))
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 3073, in iter
                  for key, pa_table in ex_iterable.iter_arrow():
                                       ~~~~~~~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2581, in _iter_arrow
                  yield from self.ex_iterable._iter_arrow()
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 558, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 440, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 345, in _generate_tables
                  raise e
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 291, in _generate_tables
                  pa_table = paj.read_json(
                      io.BytesIO(batch), read_options=paj.ReadOptions(block_size=block_size)
                  )
                File "pyarrow/_json.pyx", line 364, in pyarrow._json.read_json
                File "pyarrow/error.pxi", line 155, in pyarrow.lib.pyarrow_internal_check_status
                File "pyarrow/error.pxi", line 92, in pyarrow.lib.check_status
                  raise convert_status(status)
              pyarrow.lib.ArrowInvalid: JSON parse error: Invalid value. in row 0

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

French Speech Dataset (Sample)

Free sample recordings from Spirelight's off-the-shelf French speech dataset. 1,131 hours of unscripted French recorded in January and February 2026, mostly by speakers in Algeria, Madagascar, Morocco, Rwanda and Benin: 1,012 hours of two-person conversation, each speaker on their own channel, and 118 hours of single-speaker recordings, with machine transcripts and word-level timestamps. Download access within 24 hours of a signed order.

Full dataset, pricing and quotes: spirelight.ai/datasets/french-france

What is in this sample

3 complete recordings, 78 minutes in total. The audio is unedited and compressed losslessly to FLAC here; buyers receive WAV plus one file per speaker for conversations. Each recording has its uncorrected machine transcript as TXT and JSON.

File Type Length Audio
conversation/french-conversation-01.flac Conversation 31 min 48 kHz, 16-bit, stereo
monologue/french-monologue-01.flac Monologue 21 min 48 kHz, 24-bit, mono
monologue/french-monologue-02.flac Monologue 25 min 48 kHz, 24-bit, mono

Each audio folder has a metadata.csv with the full transcript text, so the dataset viewer and datasets can read it directly. Word-level timestamps are in the JSON files under transcripts/.

from datasets import load_dataset

ds = load_dataset("SpeechDataAI/french-speech-dataset", "conversation", split="train")
print(ds[0]["text"][:300])

The full dataset

Recording type Recordings Hours Average length
Conversation, business topics 919 505.9 33 min
Conversation, open topics 936 506.4 32 min
Monologue 221 109.4 30 min
Single speaker, type not labelled 12 8.9 44 min

Recordings from France make up 27 hours. If you need speakers from France, Belgium, Switzerland or Canada, we can record them to order.

Speaker location Hours
Algeria 487
Madagascar 202
Morocco 139
Rwanda 105
Benin 95
France 27
Côte d'Ivoire 25
Senegal 13
Burundi 12
Other or not stated 26

Speakers' home regions include Antananarivo, Algiers, Béjaïa, Batna and Casablanca.

  • Speakers: 82 registered contributors, 27 of whom recorded monologues. Their conversation partners were not registered, so there are more than 82 distinct voices; that number is not counted.
  • Recorded: 2026-01-24 to 2026-02-26
  • Audio: 48 kHz WAV. Conversations: stereo, one speaker per channel, plus one file per speaker. Single-speaker recordings: mono. Bit depth is 16 or 24 depending on the recording.
  • Devices and setting: 96% recorded on a laptop or PC and about 4% on a phone, in quiet indoor settings.
  • Transcripts: Machine transcripts with speaker labels, segment start and end times and word-level timestamps, as TXT and JSON, for 1,129 of the 1,131 hours. Not human-reviewed.
  • Topics: More than 1,700 distinct topic labels, from business meetings, interviews and sales to football, music, religion and daily life.
  • Licence: commercial use under a written licence; voice cloning is excluded.
  • Price: $35 per audio hour for all 1,131 hours ($39,585); a part of the dataset is quoted.
  • Delivery: download access within 24 hours of a signed order.

Good for

  • Training and evaluating French ASR (speech recognition) on unscripted speech rather than read sentences.
  • Speaker diarization: in the conversations each speaker has their own channel, so turns and overlaps can be read from the channels.
  • Voice agents and turn-taking research: real interruptions, backchannels and topic changes.

Not for voice cloning or imitating a speaker.

How the samples differ from free corpora

Common Voice and FLEURS are short sentences read aloud, and MLS is audiobooks read aloud. These recordings are unscripted conversation and monologue, 32 minutes per recording on average.

Licence for this sample

The files in this repository are free to download and use to evaluate whether the full dataset fits your project. Training on them, redistributing them and voice cloning are not permitted. See LICENSE.md. The full dataset is licensed commercially: request a quote or the full sample pack.

About

Published by SpeechData.ai, part of Spirelight (Spire Light ApS, Denmark), which records, transcribes and licenses speech data for teams building voice AI. Browse every dataset at spirelight.ai/datasets.

Downloads last month
38

Collection including SpeechDataAI/french-speech-dataset