English Speech Files

Nested
ralfherzog-20070819_vf17.zip
User: ralfherzog
Date: 8/18/2007 9:10 pm
Views: 3403
Rating: 14

Hello everyone!  After I have recorded the audio files with audacity, I have let them transcribe with DNS 9.  The following words weren't recognized correctly:

In the line 8 it was recognized "chants" instead of "chance."
In the line 9 it was recognized "Jean" instead of "Jeanne."
In the line 12 it was recognized "has on"  instead of "hastened."  Another recognition failure was that it was recognized "PR" instead of "Pierre."
In the line 16 it was recognized "Gene has" instead of "Jeanne's."
In the line 19 it was recognized "gene" instead of "Jeanne."
In the line 31 it was recognized "jean" in stead of "Jeanne."
In the line 35 was recognized "of the Day" instead of "of day."  Obviously DNS 9 adds the word "the" even if it hasn't been dictated.

Maybe you can learn something from this analysis: one problem is the recognition of the word "Jeanne".  I have to train DNS 9 to get the correct result.  
Another interesting observation is that DNS 9 obviously sometimes adds small words like "the" even if it hasn't been dictated.  So this observation should be kept in mind when developing the open source speech recognition software.  It is not only about the recognition of the speech.  The software might add some small words, even if explicitly not dictated.

Speaker Characteristics:

Gender: male;
Age range: adult;
Pronunciation dialect: General American.

Recording Information:

Microphone make: Sennheiser PC131;
Microphone type: noise canceling headset;
Audio card make: OnBoard - Mainboard: Asrock 939Dual-SATA2;
Audio card type: OnBoard - Realtek 850;
Audio Recording Software: Audacity 1.2.6;
O/S: Windows XP Professional.

File Info:

File type: wav;
Sampling rate: 48kHz;
Sample rate format: 16bit;
Number of channels: 1;
Audio Processing: n

vf17-01 Something about it seemed to fascinate him, to challenge his presence
vf17-02 Now it was missing from the wall
vf17-03 He boiled himself some coffee and sat down to wait
vf17-04 I'm going down there with you, and I'm going to fight
vf17-05 Now have you got anything to say against me, Mr Philip
vf17-06 If I meet her again I shall apologize, said Eileen
vf17-07 Below him the shadow was broken into a pool of rippling starlight
vf17-08 Only the chance sound had led him to observe them
vf17-09 Could the incident have anything to do with Jeanne and Pierre
vf17-10 There was no chance to fire without hitting him
vf17-11 There was no answer from the other side
vf17-12 Then he hastened on, as Pierre had guided him
vf17-13 With these arguments he convinced himself that he should go on alone
vf17-14 Yet, behind them there was another and more powerful motive
vf17-15 In that case he could not miss them, if he used caution
vf17-16 Before Philip could recover himself Jeanne's startled guards were upon him
vf17-17 It is the nearest refuge
vf17-18 There was pride and strength, the ring of triumph in his voice
vf17-19 The truth of it set Jeanne quivering
vf17-20 Tomorrow it will be strong enough for you to stand upon
vf17-21 You were going to leave after you saw me on the rock
vf17-22 He bit his tongue, and cursed himself at this fresh break
vf17-23 In it there was something that was almost tragedy
vf17-24 Your face is red with blood
vf17-25 Her eyes smiled truth at him as he came up the bank
vf17-26 He can care for himself
vf17-27 They will search for us between their camp and Churchill
vf17-28 Until I die, he exclaimed
vf17-29 Her beautiful hair was done up in shining coils
vf17-30 The Churchill narrowed and its current became swifter as they progressed
vf17-31 For a full half minute Jeanne looked at him without speaking
vf17-32 I want to die in it
vf17-33 Darkness hid him from Jeanne
vf17-34 And yet if she came he had no words to say
vf17-35 He heard a sound which brought him quickly into consciousness of day
vf17-36 Within himself he called it no longer his own
vf17-37 Besides, that noise makes me deaf
vf17-38 Philip looked back from the crest and saw Jeanne leaning over the canoe
vf17-39 Fifty yards ahead of her were the first of the rocks
vf17-40 There was one chance, and only one, of saving Jeanne

Copyright (C) 2007  Ralf Herzog

These files are free software; you can redistribute them and/or
modify them under the terms of the GNU General Public License
as published by the Free Software Foundation; either version 2
of the License, or (at your option) any later version.

These files are distributed in the hope that they will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
GNU General Public License for more details.

--- (Edited on 8/18/2007 9:10 pm [GMT-0500] by ralfherzog) ---

ralfherzog-20070819_vf17.zip ralfherzog-20070819_vf17.zip

Notice: many prompts in "English Speech Files" were adapted from the prompt files contained in the CMU_ARCTIC speech synthesis database, which were in turn derived from out-of-copyright texts from Project Gutenberg, by the FestVox project at the Language Technologies Institute at Carnegie Mellon University.

Re: ralfherzog-20070819_vf17.zip
User: kmaclean
Date: 8/20/2007 1:50 pm
Views: 216
Rating: 20

Hi Ralph,

thanks again for your submissions (in English and German)!

>Another interesting observation is that DNS 9 obviously sometimes adds small words like "the" even if it hasn't been dictated.  

This is very interesting and will have to be looked at in more detail when we start working on a Dialog Manager for dictation.

thanks,

Ken 

 

--- (Edited on 8/20/2007 2:50 pm [GMT-0400] by kmaclean) ---


Notice: many prompts in "English Speech Files" were adapted from the prompt files contained in the CMU_ARCTIC speech synthesis database, which were in turn derived from out-of-copyright texts from Project Gutenberg, by the FestVox project at the Language Technologies Institute at Carnegie Mellon University.

speech recognition software that adds words
User: ralfherzog
Date: 8/20/2007 6:39 pm
Views: 396
Rating: 21
Hi Ken,

Yes, I think that is interesting when the dictation software adds words that haven't been dictated.  

Three years ago, I used the speech recognition software ViaVoice version 10. ViaVoice added often words, even when I said nothing at all. But those words didn't make any sense.  It was obviously a *weakness* of ViaVoice.  

DNS 9 behaves in a different way.  It adds words only occasionally, and only if it makes sense.  So the text might have a better quality.  So this could be some kind of a grammar correction that is implemented in DNS 9.

Maybe you can learn something from those observations.  We can observe how commercial speech recognition software behaves in different situations.  And maybe we can learn from these different kinds of behaviors.

I think it is possible to learn something from the differences between ViaVoice 10 (which is a bad speech recognition software) and DNS 9 (which is a good speech recognition software).

So, to get a good free open source speech recognition software there has to be some kind of correction mode implemented.  I don't know how to achieve this goal, but let's think about it.

Greetings, Ralf

--- (Edited on 8/20/2007 6:39 pm [GMT-0500] by ralfherzog) ---


Notice: many prompts in "English Speech Files" were adapted from the prompt files contained in the CMU_ARCTIC speech synthesis database, which were in turn derived from out-of-copyright texts from Project Gutenberg, by the FestVox project at the Language Technologies Institute at Carnegie Mellon University.

PreviousNext