Flaky LibriSpeech
This dataset is constructed by training a character-based Conv1d-LSTM on train-other-500 and train-clean-360 and decoding train-clean-100 after every epoch. The intended use of this system is training active learning / utterance verification systems.
The run that resulted in this data: https://wandb.ai/stud76/ha/runs/3i0kv0vf
The code for the model and training is available in haloop commit e7b6fe7. Implementation of Bootstrapped Likelihood is available in the main branch of haloop
This dataset contains 28539 utterances. Later experiments use 28538 utterances without the first one.
Examples
;; grep -H 1183-133256-0000 *txt | awk -F'[:\t]' '{print $1, $3}'
train-clean-100.ep05.txt RIGHT THREMATIC RAMIDY RIGHT S CULTHALL BALM TWO GREAT MADICINES THESE WELL KNOW COMMEDEIAMENICINS ARE OF A HIGH ORER O EXCELLENCE OF THE GREATES FALUE PROMPT N ACTION AND WELIEF TRI ABOTTLE OF RIGHT TRUMATIC RAMITY FOR YOU ROMAT HISM
train-clean-100.ep10.txt RIGHTS THROUGH MATIC RAMETY RIGHT COULTALL BAM TWO GREAT MEDICINES THESE WELL NOW CONMEDIAM MENICINS RERE OF A HIGH ORDER OF EXCELLENCE OT OF THE GREATES VALUE PROMPT IN ACTION AND WELIEF TRY ABOTTLE OF RIHT TRUG MATCRAMITY FOR YOU RIMATISM
train-clean-100.ep20.txt RIGHT THREMATIC RAMEDY RIGHTS CATALE BAM TWO GREAT MEDICINES THESE WELL KNOW CANMEDIUM MEDICINS RE OF A HIGH ORDER OF EXCELLENCE ON OF THE GREATS VALUE PROMPT ON ACTION AND WELIEF TRY ABOTTLE OF RIGHT TRUMATACGRAMITY FOR YOUR RIMATISM
train-clean-100.ep30.txt RIGHT THREEMATIC REMEDY RIGHTS KUTARAL BALN TWO GREAT MEDICINES THESE WELL KNOW CANMEDIAMT MEDICINS ARE OF A HIGH ORDER OF EXCELLENCE ON OF THE GREATES FALUE PROMPT AND ACTION AND RELIEF TRIABOTTLE OF RIGHT RMATIC RAMETY FOR YOUR RIMATISM
train-clean-100.ep40.txt RIGHT HREMATIC REMEDY RITES CAULTARAL BALM TWO GREAT MEDICINES THESE WILL KNOW CONEDIUM MEDICINS WRE OF A HIGH ORDER OF EXCELLENCE OF O THE GREATEST VALUE PROMPTAND ACTION AND RELIEF TRIA BOTTLE OF RIGHTS THROUMATIGRAMITY FOR YOUR REMATISM
train-clean-100.ep45.txt RIGHT TO REMATIC RAMEDY WRIGHTS CAUTARAL BALM TWO GREAT MEDICINES THESE WELL KNOW CANEDIAN MEDICINES ARE OF A HIGH ORDER OF EXCELLENCE AN OF THE GREAT ST VALUE PROMPTON ACTION AND RELIEF TRYABOTTLE OF RIGHT TROMANTICRAMITY FREOR REMITISM
train-clean-100.ref.txt WRIGHT'S RHEUMATIC REMEDY WRIGHT'S CATARRHAL BALM TWO GREAT MEDICINES THESE WELL KNOWN CANADIAN MEDICINES ARE OF A HIGH ORDER OF EXCELLENCE AND OF THE GREATEST VALUE PROMPT IN ACTION AND RELIEF TRY A BOTTLE OF WRIGHT'S RHEUMATIC REMEDY FOR YOUR RHEUMATISM
corrupted-librispeech;; grep -H 5022-29411-0030 *txt | awk -F'[:\t]' '{print $1, $3}'
train-clean-100.ep05.txt E WAYS SHAE I FOLLOWED AT MY HUMBLE DISTANCE DE EXAMPLE OF SIR WALTER SCOT WHEN INQURISITIVE PEOPLE ASFD HIM OF HE WAS THE OFTHOR OF THE WAVERIN OWLS INPLAIN INGLISH I DENID ALL KNOWLEDGE OF THE STRANGER WHEREIN THE GREEN HAT
train-clean-100.ep10.txt A WAYS SHAE I FOLLOWED AT MY HUNBLE DISTANCE THE EXAMPLE OF SIR WALTER SCOT WHN INQUISITIVE PEOPLE ASKED HIM OF HE WAS THE AUTHOR OF THE WAVERLINOLS INPLAIN INGLISH I DENIED ALL KNOWLEDGE OF THE STRANGER WHERING THE GREEN HAT
train-clean-100.ep20.txt A WAS SHAKE I FOLLOWED AT MY HUMBLE DISTANCE THE EXAMPLE OF SIR WALTER SCOT WHEN INQUISITIVE PEOPLE ASKED HIM IF HE WAS THE AUTHOR OF THE WAVELINOVALS IM PLAIM ING WISH I DENIED ALL KNOWLEDGE OF THE STRANGER WHERING THE GREEN HAT
train-clean-100.ep30.txt A IS SHE I FOLLOWED AT MY HUMBLE DISTANCE THE EXAMPLE OF SIR WALTER SCOT WHEN INQUISITIVE PEOPLE ASKED HIM IF HE WAS THE AUTHOR OF THE WAVERIN OVALS IM PLAIN ENGLISH I DINIED ALL KNOWLEDGE OF THE STRANGER WERING THE GREEN HAT
train-clean-100.ep40.txt A WAYS SHA I FOLLOWED AT MY HUMBLE DISTANCE THE EXAMPLE OF SIR WALTERS SCOT WHEN INQUISITIVE PEOPLE ASKED HIM IF HE WAS THE AUTHOR OF THE WAVERLINOVELS IM PLAIN INGLISH I DENIED ALL KNOWLEDGE AS A STRANGER WERIN THE GREEN HAT
train-clean-100.ep45.txt A WIS SHA I FOLLOWED AT MY HUMBLEDISTANCE THE EXAMPLE OF SIR WALTER SCOT WHEN INQUISITIVE PEOPLE ASKED HIM AF HE WAS THE AUTHOR OF THE WAVERLY NOVALS IM PLAININGLISH I DINIED ALL KNOWLEDGE OF THE STRANGER WERIN THE GREEN HAT
train-clean-100.ref.txt WHO IS SHE I FOLLOWED AT MY HUMBLE DISTANCE THE EXAMPLE OF SIR WALTER SCOTT WHEN INQUISITIVE PEOPLE ASKED HIM IF HE WAS THE AUTHOR OF THE WAVERLEY NOVELS IN PLAIN ENGLISH I DENIED ALL KNOWLEDGE OF THE STRANGER WEARING THE GREEN HATcorrupted-librispeech;; grep -H 1578-6379-0022 *txt | awk -F'[:\t]' '{print $1, $3}'
train-clean-100.ep05.txt AND I WENT STRAIGHT INTO TE WITHOUT GOING OF STAIRS THE BUTLER TOOK UT MY SICASE AND THE MAYE PAME AN US FOR THE KEYS AS SHE COULD ANPACK THE HOUSES SIMPLY RENNING OVER WITH SERVANTS AN ONLY SCARE TO DEATH
train-clean-100.ep10.txt AND I WENT STRAIGHT INTO TE WITHOUT GOING UP STAIRS THE BUTLER TOOK UP MY SU CASE AND THE MAY CAME EN US FOR THE KEYS OFE SHE COULD INPACK THE HOUSES SIMLY RUNNING OVER WITH SERVANTS AN ONLY SCARE TO DEATH
train-clean-100.ep20.txt AND I WENT STRAIGHT INTO TE Y WITHOUT GOING UP STAIRS THE BUTLER TOOK UP MY SUCASE AND THE MADE CAME IN ASK FOR THE KEYS OS SHE COULD IMPACK THE HOUSEIS SIMPLY RENNING OVER WITH SERVANTS ON MOLY SCARC TO DEATH
train-clean-100.ep30.txt AND I WENT STRAIGHT INTO TE WITHOUT GOING UPSTAIRS THE BUTLER TOOK UP MY SUC CASE AND THE MAK CAME AND ASK D FOR THE KEYS S SHE COULD IMPACK THE HOUSEIS SIMPLY RENNING OVER WITH SERVANTS AN ONLY SCARE TO DEATHS E LITLE
train-clean-100.ep40.txt AND I WENT STRAIGHT INTO TE WITHOUT GOING UPSTAIRS THE BUTLER TOOK UP MY SU CASE AND THE MADE CAME AN ASK FOR THE KEY SH SHE COULD INPACK THE HOUSES SIMPLY RUNNING OVER WITH SERVANTS ON MOLLY SCARE TO DEATH IN ED A IESNS SAS
train-clean-100.ep45.txt AND I WENT STRAIGHT INTO TEA WITHOUT GOING UPSTAIRS THE BUTLER TOOK UP MY SU CASE AND THE MAYE CAME IN SK FOR THE KEY OE SHE COULD AIN PACK THE HOUSEIS SIMPLY RUNNING OVER WITH SERVANTS ON MOLY SCARE TO DEATH
train-clean-100.ref.txt AND I WENT STRAIGHT IN TO TEA WITHOUT GOING UPSTAIRS THE BUTLER TOOK UP MY SUIT CASE AND THE MAID CAME AND ASKED FOR THE KEY SO SHE COULD UNPACK THAT HOUSE IS SIMPLY RUNNING OVER WITH SERVANTS I'M ALWAYS SCARED TO DEATH