Datasets:
Tasks:
Automatic Speech Recognition
Formats:
soundfolder
Sub-tasks:
speaker-identification
Languages:
English
Size:
< 1K
License:
| *** | |
| *** THIS IS A SAMPLE DATABASE ONLY. THE INFORMATION CONTAINED IN | |
| *** THE REST OF THIS README APPLIES TO THE FULL DATABASE | |
| *** | |
| This is a US English conversational database, produced by Appen Butler Hill Pty. Ltd. | |
| Appen Butler Hill Pty. Ltd. owns copyright of the database. | |
| The database contains speech data recorded during 928 sessions. Each of the 928 unique speaker-pairs was recorded making conversations of approximately an average of 60 minutes | |
| Each pair of speakers recorded up to 14 conversations on different topics. Speakers were provided with a topic for each conversation. | |
| The recordings have been made using the Appen AMR smartphone app with the phone positioned between two speakers who are in the same room. | |
| The database contains approximately 1005 hours of audio data in total. | |
| The directory structure is designed to group data from each pair of speakers in a single folder. Each pair of speakers has been identified with a unique ID. This ID is also the name of each session folder. Within each session folder are the conversations made by the pair of speakers. The file naming structure indicates the session ID (which is the same as the speaker pair ID), country of origin for speaker 1 and speaker 2, and conversation ID. | |
| Directory Structure: | |
| /+-COPYRIGHT.TXT | |
| +-README.TXT | |
| +-DEMOGRAPHICS.CSV | |
| +-Transcription_Conventions_20180712.pdf | |
| +-LEXICON--------+-USE_ASR003_Lexicon.txt | |
| | +-USE_ASR003_Lexicon_README.txt | |
| | | |
| | | |
| +-AUDIO----------+-F1M1----------------+-F1M1_USA_USA_001.wav | |
| | | +-F1M1_USA_USA_003.wav | |
| | | . | |
| | | . | |
| | | +-F1M1_USA_USA_013.wav | |
| | | | |
| | | | |
| | +-F2M2----------------+-F6M6_USA_USA_001.wav | |
| | | +-F6M6_USA_USA_003.wav | |
| | | . | |
| | | . | |
| | | +-F6M6_USA_USA_011.wav | |
| | . | |
| | . | |
| | . | |
| | +-M25F25--------------+-M25F25_USA_USA_001.wav | |
| | +-M25F25_USA_USA_002.wav | |
| | . | |
| | . | |
| | +-M25F25_USA_USA_005.wav | |
| | | |
| | | |
| | | |
| +-TRANSCRIPTION_AUTO_SEGMENTED---------+-F1M1-----------------+-F1M1_USA_USA_001.txt | |
| | | +-F1M1_USA_USA_003.txt | |
| | | . | |
| | | . | |
| | | +-F1M1_USA_USA_013.txt | |
| | | | |
| | | | |
| | +-F2M2-----------------+-F6M6_USA_USA_001.txt | |
| | | +-F6M6_USA_USA_003.txt | |
| | | . | |
| | | . | |
| | | +-F6M6_USA_USA_011.txt | |
| | . | |
| | . | |
| | . | |
| | +-M25F25---------------+-M25F25_USA_USA_001.txt | |
| | +-M25F25_USA_USA_002.txt | |
| | . | |
| | . | |
| | +-M25F25_USA_USA_005.txt | |
| | | |
| | | |
| +-TRANSCRIPTION_SEGMENTED_TO_SENTENCES-+-F1M1-----------------+-F1M1_USA_USA_001.txt | |
| | +-F1M1_USA_USA_003.txt | |
| | . | |
| | . | |
| | +-F1M1_USA_USA_013.txt | |
| | | |
| | | |
| +-F2M2-----------------+-F6M6_USA_USA_001.txt | |
| | +-F6M6_USA_USA_003.txt | |
| | . | |
| | . | |
| | +-F6M6_USA_USA_011.txt | |
| . | |
| . | |
| . | |
| +-M25F25---------------+-M25F25_USA_USA_001.txt | |
| +-M25F25_USA_USA_002.txt | |
| . | |
| . | |
| +-M25F25_USA_USA_005.txt | |
| COPYRIGHT.TXT is a copyright document in ASCII format. | |
| README.TXT is this file. It is an ASCII text file that lists all files of the database. | |
| DEMOGRAPHICS.CSV is an Excel file that contains the following fields: | |
| - filename | |
| - domains-topics | |
| - device-type | |
| - gender01 | |
| - gender02 | |
| - age-group01 | |
| - age-group01 | |
| - country-of-residence01 | |
| - country-of-residence02 | |
| - country-of-origin01 | |
| - country-of-origin02 | |
| CONVERSATION TOPICS: | |
| The conversations were spontaneous and were on a variety of generic topics (e.g. news, travel, study etc.). The topics can be found in the Demographics file. | |
| /AUDIO contains all audio data from the 928 session. Audio format is WAVE audio, Microsoft PCM, 16 bit, mono 16000 Hz | |
| /TRANSCRIPTION_SEGMENTED_TO_SENTENCES contains all transcription data. This data was segmented automatically with the segment boundaries determined by silence detection | |
| /TRANSCRIPTION_AUTO_SEGMENTED contains the same transcription data as the folder above, however, the data was segmented manually in such a way that the segment boundaries are time aligned to sentence/meaningful sematnic unit boundaries. | |
| The Audio and Transcription filenames use the following template: | |
| <Speaker1ID><Speaker2ID>_<country1>_<country2>_<recordingID>.wav | |
| Where: | |
| <Speaker1ID> = unique character speaker code, where the first character represents the speaker gender (M=Male, F=Female) and the second character corresponds to the speaker pair number. | |
| <Speaker2ID> = unique character speaker code, where the first character represents the speaker gender (M=Male, F=Female) and the second character corresponds to the speaker pair number. | |
| <country1> = ISO code for the country of origin of Speaker1 | |
| <country2> = ISO code for the country of origin of Speaker2 | |
| <RecordingID> = session ID | |
| E.g. F1M1_USA_USA_001.wav | |
| PS. There are some sessions (22 sessions) that have no demographics. The filename of these sessions, start with "None... | |
| TRANSCRIPTION TAGS: | |
| - Speech event tagging | |
| Timestamp [12.345] | |
| Speaker IDs <speaker_1> <speaker_2> <speaker_3> <speaker_4> | |
| Hesitation <hes> | |
| Mispronunciation *word* | |
| Fragments - | |
| Restart -- | |
| Truncations ~ | |
| Foreign words and codeswitching <foreign> | |
| Unintelligible speech (()) | |
| - Non-speech/acoustic event tagging | |
| Background sound <sta> | |
| Overlap <overlap> | |
| No speech <no-speech> | |
| - Speaker-originated foreground noise | |
| <lipsmack> (lip smacks, tongue clicks) | |
| <breath> (inhalation and exhalation between words, yawning) | |
| <cough> (coughing, throat clearing, sneezing) | |
| <laughter> (Laughing, chuckling) | |
| - Other source foreground noise | |
| <click> (machine or recording click) | |
| <int> (any other intermittent foreground noise) | |
| /LEXICON: contains pronunciations for all words transcribed in the database and stats report. | |
| STATISTICS FOR LEXICON: | |
| ### Totals ### | |
| Total headwords = 52586 | |
| Total pronunciations = 64209 | |
| ### Variant Count ### | |
| Words with 1 variants: 42956 | |
| Words with 2 variants: 8304 | |
| Words with 3 variants: 721 | |
| Words with 4 variants: 567 | |
| Words with 5 variants: 16 | |
| Words with 6 variants: 21 | |
| Words with 8 variants: 1 | |
| ### Phoneme Counts ### | |
| Unique Phones: 46 / 46 | |
| p 11188 | |
| b 8776 | |
| t 24204 | |
| d 16359 | |
| k 18809 | |
| g 5106 | |
| tS 2321 | |
| dZ 3532 | |
| f 6508 | |
| v 5407 | |
| T 1342 | |
| D 334 | |
| s 24304 | |
| z 15304 | |
| S 4359 | |
| Z 319 | |
| h 3174 | |
| m 12388 | |
| n 25543 | |
| N 6252 | |
| 4 4371 | |
| r 23819 | |
| l 21324 | |
| j 2738 | |
| w 3892 | |
| V 4724 | |
| A 10912 | |
| { 10268 | |
| E 13079 | |
| 3` 2557 | |
| I 25433 | |
| i 18873 | |
| O 4176 | |
| U 904 | |
| u 4913 | |
| @ 31515 | |
| @` 7655 | |
| aI 6011 | |
| eI 8347 | |
| OI 547 | |
| aU 1331 | |
| oU 6851 | |
| " 68284 | |
| % 2524 | |
| . 89732 | |
| # 4155 | |