robertanto / aws-lex-noisy-spoken-language-understanding Goto Github PK

This repository contains dataset for evaluating spoken language understanding system with audio input. The utterances, intent and slot labeling in the dataset comes from SNIPS dataset published by the Snips team:

• Coucke A. et al., "Snips Voice Platform: an embedded Spoken Language Understanding system for private-by-design voice interfaces." 2018
• github repo

Here we extend the natural language understanding (NLU) dataset with audio by synthesizing the utterance with different voices provided by Amazon Polly. More information about the voices can be found here.

Due to the size of the dataset, we host the audio files on s3. After you download and unzip the file, you will get a folder with following structure

audio_slu/
│
├── audio_$voice_id_1/
│   └── snips
│       ├── $utterance_id_1.mp3
│       ├── $utterance_id_2.mp3
│       ├── ...
│ 
├── audio_$voice_id_2/
│   └── snips
│       ├── $utterance_id_1.mp3
│       ├── $utterance_id_2.mp3
│       ├── ...
│
├── ...

voice_id is the Amazon polly voice used to synthesize the audio as listed in the doc. utterance_id is the utterance id can be used to relate the audio with its transcript and NLU annotation.

The transcript and NLU annotation for each utterance can be found in the folder. u_id in the file can be used to relate NLU annotation with audio.

See CONTRIBUTING for more information.

See the LICENSE file for more information.