Dataset Viewer
The dataset viewer is not available for this subset.
Cannot get the split names for the config 'default' of the dataset.
Exception: SplitsNotFoundError
Message: The split names could not be parsed from the dataset config.
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
for split_generator in builder._split_generators(
~~~~~~~~~~~~~~~~~~~~~~~~~^
StreamingDownloadManager(base_path=builder.base_path, download_config=download_config)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/webdataset/webdataset.py", line 81, in _split_generators
first_examples = list(islice(pipeline, self.NUM_EXAMPLES_FOR_FEATURES_INFERENCE))
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/webdataset/webdataset.py", line 32, in _get_pipeline_from_tar
fs: fsspec.AbstractFileSystem = fsspec.filesystem("memory")
~~~~~~~~~~~~~~~~~^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/fsspec/registry.py", line 302, in filesystem
cls = get_filesystem_class(protocol)
File "/usr/local/lib/python3.14/site-packages/fsspec/registry.py", line 239, in get_filesystem_class
raise ValueError(f"Protocol not known: {protocol}")
ValueError: Protocol not known: memory
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/split_names.py", line 71, in compute_split_names_from_streaming_response
for split in get_dataset_split_names(
~~~~~~~~~~~~~~~~~~~~~~~^
path=dataset,
^^^^^^^^^^^^^
config_name=config,
^^^^^^^^^^^^^^^^^^^
token=hf_token,
^^^^^^^^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
info = get_dataset_config_info(
path,
...<6 lines>...
**config_kwargs,
)
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.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.
Bulgarian TTS Dataset (1300 Hours)
This is a large-scale, high-quality Bulgarian Text-to-Speech (TTS) dataset containing approximately 1300 hours of transcribed audio.
Dataset Curation & Filtering
The original raw dataset consisted of 3000 hours of audio. To ensure the highest quality for training TTS models (and specifically for optimizing model context windows), a rigorous filtering and cleaning pipeline was applied. The final dataset is 1300 hours, with the following filters applied:
- Audio Duration Filter: Only segments between 2 and 12 seconds were kept. This enforces a small context window (ideal for models with 512-1024 token contexts, enabling efficient training of small parameters models).
- Short Text Filter: Removed any segments containing 0 or 1 words (often sighs, breaths, or noise).
- Number Normalization: Converted digits (1, 2, 3...) and Latin symbols into their Cyrillic spoken word equivalents to avoid OOV (Out-Of-Vocabulary) errors during TTS inference.
- Hapax Legomena Filter: Removed segments containing words that appear only once in the entire 3000-hour dataset (Hapax Legomena). These were statistically proven to be OCR transcription errors or misspellings.
- Characters Per Second (CPS) Filter: Removed segments with unnatural speaking rates (too fast or too slow), ensuring a natural and consistent speech pacing across the dataset.
- Repetition Filter: Removed segments containing 3 or more consecutive identical words (which are usually artifacts or stutters).
Format
- Audio: FLAC format (converted from WAV)
- Transcriptions: Provided in
metadata.jsonl - Speaker Anonymization: The original speaker identities have been anonymized (e.g.,
Speaker_001,Speaker_002) to prevent unauthorized voice cloning of specific actors, while still preserving speaker identity consistency for Multi-Speaker TTS training.
Usage
This dataset is optimized for training modern Multi-Speaker TTS architectures (e.g., VALL-E, XTTS, FastSpeech2) due to the strict 12-second segment limit and thoroughly normalized text.
- Downloads last month
- 65