mindspore.dataset.TedliumDataset

class mindspore.dataset.TedliumDataset(dataset_dir, release, usage=None, extensions=None, num_samples=None, num_parallel_workers=None, shuffle=None, sampler=None, num_shards=None, shard_id=None, cache=None)[source]

Tedlium dataset. The columns of generated dataset depend on the source SPH files and the corresponding STM files.

The generated dataset has six columns [waveform, sample_rate, transcript, talk_id, speaker_id, identifier] .

The data type of column waveform is float32, the data type of column sample_rate is int32, and the data type of columns transcript , talk_id , speaker_id and identifier is string.

Parameters

dataset_dir (str) – Path to the root directory that contains the dataset.
release (str) – Release of the dataset, can be 'release1', 'release2', 'release3'.
usage (str, optional) – Usage of this dataset. For release1 or release2, can be 'train', 'test', 'dev' or 'all'. 'train' will read from train samples, 'test' will read from test samples, 'dev' will read from dev samples, 'all' will read from all samples. For release3, can only be 'all', it will read from data samples. Default: None , all samples.
extensions (str, optional) – Extensions of the SPH files, only '.sph' is valid. Default: None , set to ".sph".
num_samples (int, optional) – The number of audio samples to be included in the dataset. Default: None , all samples.
num_parallel_workers (int, optional) – Number of worker threads to read the data. Default: None , will use global default workers(8), it can be set by mindspore.dataset.config.set_num_parallel_workers() .
shuffle (bool, optional) – Whether to perform shuffle on the dataset. Default: None , expected order behavior shown in the table below.
sampler (Sampler, optional) – Object used to choose samples from the dataset. Default: None , expected order behavior shown in the table below.
num_shards (int, optional) – Number of shards that the dataset will be divided into. Default: None . When this argument is specified, num_samples reflects the maximum sample number of per shard.
shard_id (int, optional) – The shard ID within num_shards . Default: None . This argument can only be specified when num_shards is also specified.
cache (DatasetCache, optional) – Use tensor caching service to speed up dataset processing. More details: Single-Node Data Cache . Default: None , which means no cache is used.

Raises

RuntimeError – If dataset_dir does not contain stm files.
RuntimeError – If sampler and shuffle are specified at the same time.
RuntimeError – If sampler and num_shards/shard_id are specified at the same time.
RuntimeError – If num_shards is specified but shard_id is None.
RuntimeError – If shard_id is specified but num_shards is None.
ValueError – If num_parallel_workers exceeds the max thread numbers.
ValueError – If shard_id is not in range of [0, num_shards ).

Tutorial Examples:

Load & Process Data With Dataset Pipeline

Note

The parameters num_samples , shuffle , num_shards , shard_id can be used to control the sampler used in the dataset, and their effects when combined with parameter sampler are as follows.

Sampler obtained by different combinations of parameters sampler and num_samples , shuffle , num_shards , shard_id
Parameter sampler	Parameter num_shards / shard_id	Parameter shuffle	Parameter num_samples	Sampler Used
mindspore.dataset.Sampler type	None	None	None	sampler
numpy.ndarray,list,tuple,int type	/	/	num_samples	SubsetSampler(indices = sampler , num_samples = num_samples )
iterable type	/	/	num_samples	IterSampler(sampler = sampler , num_samples = num_samples )
None	num_shards / shard_id	None / True	num_samples	DistributedSampler(num_shards = num_shards , shard_id = shard_id , shuffle = True , num_samples = num_samples )
None	num_shards / shard_id	False	num_samples	DistributedSampler(num_shards = num_shards , shard_id = shard_id , shuffle = False , num_samples = num_samples )
None	None	None / True	None	RandomSampler(num_samples = num_samples )
None	None	None / True	num_samples	RandomSampler(replacement = True , num_samples = num_samples )
None	None	False	num_samples	SequentialSampler(num_samples = num_samples )

Examples

>>> import mindspore.dataset as ds
>>> # 1) Get all train samples from TEDLIUM_release1 dataset in sequence.
>>> dataset = ds.TedliumDataset(dataset_dir="/path/to/tedlium1_dataset_directory",
...                             release="release1", shuffle=False)
>>>
>>> # 2) Randomly select 10 samples from TEDLIUM_release2 dataset.
>>> dataset = ds.TedliumDataset(dataset_dir="/path/to/tedlium2_dataset_directory",
...                             release="release2", num_samples=10, shuffle=True)
>>>
>>> # 3) Get samples from TEDLIUM_release-3 dataset for shard 0 in a 2-way distributed training.
>>> dataset = ds.TedliumDataset(dataset_dir="/path/to/tedlium3_dataset_directory",
...                             release="release3", num_shards=2, shard_id=0)
>>>
>>> # In TEDLIUM dataset, each dictionary has keys : waveform, sample_rate, transcript, talk_id,
>>> # speaker_id and identifier.

About TEDLIUM_release1 dataset:

The TED-LIUM corpus is English-language TED talks, with transcriptions, sampled at 16kHz. It contains about 118 hours of speech.

About TEDLIUM_release2 dataset:

This is the TED-LIUM corpus release 2, licensed under Creative Commons BY-NC-ND 3.0. All talks and text are property of TED Conferences LLC. The TED-LIUM corpus was made from audio talks and their transcriptions available on the TED website. We have prepared and filtered these data in order to train acoustic models to participate to the International Workshop on Spoken Language Translation 2011 (the LIUM English/French SLT system reached the first rank in the SLT task).

About TEDLIUM_release-3 dataset:

This is the TED-LIUM corpus release 3, licensed under Creative Commons BY-NC-ND 3.0. All talks and text are property of TED Conferences LLC. This new TED-LIUM release was made through a collaboration between the Ubiqus company and the LIUM (University of Le Mans, France).

You can unzip the dataset files into the following directory structure and read by MindSpore's API.

The structure of TEDLIUM release2 is the same as TEDLIUM release1, only the data is different.

.
└──TEDLIUM_release1
    └── dev
        ├── sph
            ├── AlGore_2009.sph
            ├── BarrySchwartz_2005G.sph
        ├── stm
            ├── AlGore_2009.stm
            ├── BarrySchwartz_2005G.stm
    └── test
        ├── sph
            ├── AimeeMullins_2009P.sph
            ├── BillGates_2010.sph
        ├── stm
            ├── AimeeMullins_2009P.stm
            ├── BillGates_2010.stm
    └── train
        ├── sph
            ├── AaronHuey_2010X.sph
            ├── AdamGrosser_2007.sph
        ├── stm
            ├── AaronHuey_2010X.stm
            ├── AdamGrosser_2007.stm
    └── readme
    └── TEDLIUM.150k.dic

The directory structure of TEDLIUM release3 is slightly different.

.
└──TEDLIUM_release-3
    └── data
        ├── ctl
        ├── sph
            ├── 911Mothers_2010W.sph
            ├── AalaElKhani.sph
        ├── stm
            ├── 911Mothers_2010W.stm
            ├── AalaElKhani.stm
    └── doc
    └── legacy
    └── LM
    └── speaker-adaptation
    └── readme
    └── TEDLIUM.150k.dic

Citation:

@article{
  title={TED-LIUM: an automatic speech recognition dedicated corpus},
  author={A. Rousseau, P. Deléglise, Y. Estève},
  journal={Proceedings of the Eighth International Conference on Language Resources and Evaluation (LREC'12)},
  year={May 2012},
  biburl={https://www.openslr.org/7/}
}

@article{
  title={Enhancing the TED-LIUM Corpus with Selected Data for Language Modeling and More TED Talks},
  author={A. Rousseau, P. Deléglise, and Y. Estève},
  journal={Proceedings of the Eighth International Conference on Language Resources and Evaluation (LREC'12)},
  year={May 2014},
  biburl={https://www.openslr.org/19/}
}

@article{
  title={TED-LIUM 3: twice as much data and corpus repartition for experiments on speaker adaptation},
  author={François Hernandez, Vincent Nguyen, Sahar Ghannay, Natalia Tomashenko, and Yannick Estève},
  journal={the 20th International Conference on Speech and Computer (SPECOM 2018)},
  year={September 2018},
  biburl={https://www.openslr.org/51/}
}

Pre-processing Operation

`mindspore.dataset.Dataset.apply`	Apply a function in this dataset.
`mindspore.dataset.Dataset.concat`	Concatenate the dataset objects in the input list.
`mindspore.dataset.Dataset.filter`	Filter dataset by prediction.
`mindspore.dataset.Dataset.flat_map`	Map func to each row in dataset and flatten the result.
`mindspore.dataset.Dataset.map`	Apply each operation in operations to this dataset.
`mindspore.dataset.Dataset.project`	The specified columns will be selected from the dataset and passed into the pipeline with the order specified.
`mindspore.dataset.Dataset.rename`	Rename the columns in input datasets.
`mindspore.dataset.Dataset.repeat`	Repeat this dataset count times.
`mindspore.dataset.Dataset.reset`	Reset the dataset for next epoch.
`mindspore.dataset.Dataset.save`	Save the dynamic data processed by the dataset pipeline in common dataset format.
`mindspore.dataset.Dataset.shuffle`	Shuffle the dataset by creating a cache with the size of buffer_size .
`mindspore.dataset.Dataset.skip`	Skip the first N elements of this dataset.
`mindspore.dataset.Dataset.split`	Split the dataset into smaller, non-overlapping datasets.
`mindspore.dataset.Dataset.take`	Take the first specified number of samples from the dataset.
`mindspore.dataset.Dataset.zip`	Zip the datasets in the sense of input tuple of datasets.

Batch

`mindspore.dataset.Dataset.batch`	Combine batch_size number of consecutive rows into batch which apply per_batch_map to the samples first.
`mindspore.dataset.Dataset.bucket_batch_by_length`	Bucket elements according to their lengths.
`mindspore.dataset.Dataset.padded_batch`	Combine batch_size number of consecutive rows into batch which apply pad_info to the samples first.

Iterator

`mindspore.dataset.Dataset.create_dict_iterator`	Create an iterator over the dataset that yields samples of type dict, while the key is the column name and the value is the data.
`mindspore.dataset.Dataset.create_tuple_iterator`	Create an iterator over the dataset that yields samples of type list, whose elements are the data for each column.

Attribute

`mindspore.dataset.Dataset.get_batch_size`	Return the size of batch.
`mindspore.dataset.Dataset.get_class_indexing`	Get the mapping dictionary from category names to category indexes.
`mindspore.dataset.Dataset.get_col_names`	Return the names of the columns in dataset.
`mindspore.dataset.Dataset.get_dataset_size`	Return the number of batches in an epoch.
`mindspore.dataset.Dataset.get_repeat_count`	Get the replication times in RepeatDataset.
`mindspore.dataset.Dataset.input_indexs`	Get the column index, which represents the corresponding relationship between the data column order and the network when using the sink mode.
`mindspore.dataset.Dataset.num_classes`	Get the number of classes in a dataset.
`mindspore.dataset.Dataset.output_shapes`	Get the shapes of output data.
`mindspore.dataset.Dataset.output_types`	Get the types of output data.

Apply Sampler

`mindspore.dataset.MappableDataset.add_sampler`	Add a child sampler for the current dataset.
`mindspore.dataset.MappableDataset.use_sampler`	Replace the last child sampler of the current dataset, remaining the parent sampler unchanged.

Others

`mindspore.dataset.Dataset.sync_update`	Release a blocking condition and trigger callback with given data.
`mindspore.dataset.Dataset.sync_wait`	Add a blocking condition to the input Dataset and a synchronize action will be applied.
`mindspore.dataset.Dataset.to_json`	Serialize a pipeline into JSON string and dump into file if filename is provided.