http://www.voxforge.org/home/forums/message-boards/audio-discussions http://www.voxforge.org/home/forums/message-boards/audio-discussions/a-quicker-approach-to-building-a-corpora Would it not be quicker, and less fatiguing on the user, to automatically transliterate spoken audo submissions (e.g. Librivox readings) using a best-of-breed speech recognizer (e.g. Dragon), and then have a 2-stage proofing process where any mistakes can be quickly corrected by volunteers? I believe this approach could significantly increase the transliterated words per volunteer hour that you are currently achieving, and more importantly, make the process far more enjoyable so that you can build long term relationships with volunteers, and they can continue to drive this project towards meeting its goals, rather than have them take a short term interest only. This is exactly the model that was used by Distributed Proofreaders to revolutionize the speed at which books could be ASCIIfied for submission into Project Gutenberg. That project has many parallels with what you are doing, and their process has some significant advantages over what is being used here: the process is enjoyable because there aren't too many mistakes to correct, and so you actually get to enjoy reading books at the same time -- instead, the voxforge proccess is quickly fatiguing. you can choose which books to work on which means you get to read bits of something you may have actually been interested in anyway. they have a very prominent display showing how many books (in your case transliterated hours) that have been achieved so far, and the history of that progress, and this acts as a great motivator to do more work. since their process de-couples authors from proof readers (or proof listeners in your case ;-) it allows interested parties to make long term commitments to the project that is not coupled to the optimal number of minutes needed from each author. The perfect interface I can imagine for this would let the user listen to audio in real time, while displaying sub-titles of the corresponding transliterated text, so that the user could pause the audio each time they see an error, and fix the current sub-title. I believe this should be possible by automatically dividing the audio up into sentence sized chunks before queeing it for transliteration. Another way to make this process even smoother and more enjoyable (more listening/less fixing) for the user would be to only add one chapter at a time, and then to feed the corrected audio back into the next round of training so that the next chapter required far less fix-up, and so on -- this might not be possible with Dragon, but should be with OSS recognizers, and so perhaps these would be better for subsequent chapters. Anyway, just my 2 cents! Thanks for all the good work, and I look forward to the day when high-quality speech recognizers are a part of every OSS desktop. Regards, Dominic Chambers. http://www.voxforge.org/home/forums/message-boards/audio-discussions/a-quicker-approach-to-building-a-corpora Sun, 27 Apr 2008 16:43:56 -0500 http://www.voxforge.org/home/forums/message-boards/audio-discussions/sequitur-g2p Sequitur G2P is a GPL, trainable Grapheme-to-Phoneme converter (i.e. automatically figures out the pronunciation of new words that are not in your pronunciation dictionary). From their web site: Sequitur G2P is a data-driven grapheme-to-phoneme converter developed at RWTH Aachen University - Department of Computer Science by Maximilian Bisani. The method used in this software is described inM. Bisani and H. Ney: "Joint-Sequence Models for Grapheme-to-Phoneme Conversion". Submitted for publication in Speech CommunicationAnyone used this software or familiar with the approach? How is this different (if at all) from rule-based TTS Text-to-phoneme approaches (using Festival or ESpeak)? thanks, Ken http://www.voxforge.org/home/forums/message-boards/audio-discussions/sequitur-g2p Wed, 16 Apr 2008 20:31:02 -0500 http://www.voxforge.org/home/forums/message-boards/audio-discussions/harnessing-the-self-interest-of-those-training-speaker-dependent-models Just a thought: I was reading the information about speaker dependent and speaker independent models on: http://www.voxforge.org/home/dev and it occurred to me that people who want to train the model to better recognise their voices are prime donators. If an interface collects the necessary samples to train the model to an individual's voice, the hard part is already done and a large number would likely submit the samples if asked. I realise that this isn't immediately useful, but in the future, the idea is that speech-recognition/desktop-control applications will be derived from this project. A person installing a speech-recognition program is likely to expect to spend a decent amount of time (10 minutes? 30?) training it to their voice. It would be worth keeping in mind that we want to collect the raw audio in a useful format and ask the user to submit that to Voxforge http://www.voxforge.org/home/forums/message-boards/audio-discussions/harnessing-the-self-interest-of-those-training-speaker-dependent-models Thu, 10 Apr 2008 04:33:03 -0500 http://www.voxforge.org/home/forums/message-boards/audio-discussions/using-alternate-lexicons I have successfully used the tutorial and howto with a couple of grammars now, and got thinking about alternate lexicons. The HTKBook mentions a list called BEEP which might be more suitable for my speech patterns given that the source is UK, so I downloaded the list and aborbed it into my database. I see that there are differences, including the fact that some of the phonemes are different, the voxforge lexicon knows about 'el' and 'en' but BEEP does not, and BEEP knows about 'ea', 'ia', 'oh', and 'ua' which are foreign to the voxforge list. My question is whether there are any gotchas to look out for in using "foreign" lexicons with the processes admirably laid out by voxforge processes? I'm only using my own voice for specialist grammars right now, and building from scratch. http://www.voxforge.org/home/forums/message-boards/audio-discussions/using-alternate-lexicons Fri, 14 Mar 2008 12:21:43 -0500 http://www.voxforge.org/home/forums/message-boards/audio-discussions/recognizing-the-word-computer In an earlier thread Ken noted that there was an issue with the word computer since there is an unrecognized triphone involved. I have now tripped over this stone myself, and am a bit puzzled. Computer is in the lexicon, and I have created my own audio samples (72 samples with a good sprinkling of out of vocabulary material) with the intention that the grammar will respond to my own voice. But step 4 still complains that 'computer' is not in the dictionary, even though it is in the lexicon, which I guess is different. Can anyone suggest what link I am missing here? http://www.voxforge.org/home/forums/message-boards/audio-discussions/recognizing-the-word-computer Tue, 11 Mar 2008 15:06:38 -0500 http://www.voxforge.org/home/forums/message-boards/audio-discussions/designing-grammars Has any work been done on manipulating grammars and prompt lists with databases? Seems like an ideal environment in which to test phonetic balance, adequate coverage of words in prompts, suggesting extra words to improve phonetic balance, etc. http://www.voxforge.org/home/forums/message-boards/audio-discussions/designing-grammars Mon, 10 Mar 2008 08:25:02 -0500 http://www.voxforge.org/home/forums/message-boards/audio-discussions/playback-of-recorded-prompts-fails A new Dutch 'submitter' tried to playback the prompts he recorded using the Java submission app. He got this error: unable to open the line: javax.sound.sampled.LineUnavailableException: Audio Device Unavailable He posted a screenshot here: http://forum.ubuntu-nl.org/message/209808#p209808 http://www.voxforge.org/home/forums/message-boards/audio-discussions/playback-of-recorded-prompts-fails Fri, 15 Feb 2008 11:22:49 -0600 http://www.voxforge.org/home/forums/message-boards/audio-discussions/dvd-closed-captioning-as-a-source-of-speech email from bilal ghalib: Hey guys! What a sweet project you have, I actually stumbled across it while trying to see if someone has already implemented an idea I had. I'll suggest this to you: DVD closed captioning, I have found a method to extract it and the times they happen and use this along with audio extracted 9000 hours of DVD audio/text is extracted each year, you not only get text/speech correlation, you get the times as well. What do you say? http://www.voxforge.org/home/forums/message-boards/audio-discussions/dvd-closed-captioning-as-a-source-of-speech Mon, 04 Feb 2008 12:31:10 -0600 http://www.voxforge.org/home/forums/message-boards/audio-discussions/read-the-same-things Is having the same text read by two people more or less useful than having recordings of two independent texts of similar complexity and length? http://www.voxforge.org/home/forums/message-boards/audio-discussions/read-the-same-things Tue, 27 Nov 2007 23:05:03 -0600 http://www.voxforge.org/home/forums/message-boards/audio-discussions/acoustic-model-for-mobile-devices Hi all! I have just started with that of SR. I was thinking of programing (sphinx) a small demo app for a pda or so. At this point I wonder why are there no acoustic samples for such situations?? The less the noise present in the samples the better the recognition results or it's advisable to include audio with 'normal' (for the target situation) ??? Thanks!! http://www.voxforge.org/home/forums/message-boards/audio-discussions/acoustic-model-for-mobile-devices Tue, 27 Nov 2007 04:47:19 -0600 http://www.voxforge.org/home/forums/message-boards/audio-discussions/new-speech-submission-application-is-live The new Speech Submission Application (Java applet) is now live. Users no longer need to register with VoxForge to contribute speech. You just need a current version of the Java Run-time Environment (1.5 or 1.6) on your computer. Instructions for installing Java are provided in the Java Troubleshooting Guide. If you have Java installed, the Java Run-time Environment on your http://www.voxforge.org/home/forums/message-boards/audio-discussions/new-speech-submission-application-is-live Fri, 12 Oct 2007 08:55:45 -0500 http://www.voxforge.org/home/forums/message-boards/audio-discussions/dc-offset-is-what-can-cause-background-hum-in-your-recordings MojoMove Voxcast #1 contains an excellent discussion (by Ticktockman and Robert) on DC Offset in your recordings: what it is and how to remove it. You can see the effect of DC Offset when you are recording audio, and the waveform is not correctly centered around the mid point line in an Audacity track (i.e. the zero volt axis). It usually manifests itself as a low rumbling sound in the recording. This can become a big problem if you don't record with a high enough level, and then try to normalize the audio to make the speech louder - because the rumbling noise also gets louder and can drown out your speech. Although VoxForge prefers audio submissions without any noise reduction (in order to get speech from as many different "natural environments" as possible), we will gladly accept any transcribed speech recordings. thanks, Ken http://www.voxforge.org/home/forums/message-boards/audio-discussions/dc-offset-is-what-can-cause-background-hum-in-your-recordings Mon, 08 Oct 2007 14:36:01 -0500 http://www.voxforge.org/home/forums/message-boards/audio-discussions/microphone-questions Cross posted from http://www.voxforge.org/home/forums/message-boards/audio-discussions/microphone-questions Mon, 24 Sep 2007 12:18:12 -0500 http://www.voxforge.org/home/forums/message-boards/audio-discussions/flash-recorder Hi Webmaster, Voxforge rocks!!! We have put up a flash based recorder on our website. To see it, please go to http://emandi.mla.iitk.ac.in:9000/kisanblog/loudblog/index.php and enter guest/guest as login/password You can then record files in the flash recorder. As has been previously discussed on these forums, the voxforge project needs something like that. I offer to provide you with the source code and integrate it into the voxforge site. Please contact me at abhishek[dot]singh[at]simmortel[dot]com Cheers! Abhishek. http://www.voxforge.org/home/forums/message-boards/audio-discussions/flash-recorder Sun, 23 Sep 2007 01:03:08 -0500 http://www.voxforge.org/home/forums/message-boards/audio-discussions/free-long-distance-for-telephone-speech-submission Found a site that provides fee long-distance calls called ViaTalk Free Connect in the US. The give you 10 minutes of free long-distance talk time. I've posted this information on the Telephone Speech Submission howto. Does anyone know of any other similar services in the US or elsewhere? thanks, Ken http://www.voxforge.org/home/forums/message-boards/audio-discussions/free-long-distance-for-telephone-speech-submission Tue, 18 Sep 2007 10:05:37 -0500 http://www.voxforge.org/home/forums/message-boards/audio-discussions/windows-vs-linux-audio-quality-differences Cross-posted from a post by ralfherzog (in the submissions forum): http://www.voxforge.org/home/forums/message-boards/audio-discussions/windows-vs-linux-audio-quality-differences Fri, 17 Aug 2007 07:09:04 -0500 http://www.voxforge.org/home/forums/message-boards/audio-discussions/pci-sound-card-recommendations This is a cross post from the Downloads forum (see this link). Ralph was looking for recommendations for PCI sound cards: http://www.voxforge.org/home/forums/message-boards/audio-discussions/pci-sound-card-recommendations Mon, 13 Aug 2007 09:28:41 -0500 http://www.voxforge.org/home/forums/message-boards/audio-discussions/speech-recognition-on-mpeg/audio-encoded-files The approach VoxForge has taken in processing LibriVox audiobooks is to ask LibriVox users to submit their wav files to VoxForge before they compress them to mp3 format (see the uploads page). We've also done some tests to convert mp3 speech files to wav format and training acoustic models from the wav files, and the results look promising (see the Convert Audio to MP3 and Compare Results with Original Wav link). I recently found a patent that trains acoustic models using mp3 audio directly (i.e. there is no requirement for conversion to an intermediate wav file before training acoustic models from the mp3 audio). They showed a novel(?) way of indexing videos by training acoustic models be using mp3 audio track on a video (not sure how they filter out music or other non-speech noise...). They used the HTK toolkit for this approach. Here is the abstract of the patent: http://www.voxforge.org/home/forums/message-boards/audio-discussions/speech-recognition-on-mpeg/audio-encoded-files Tue, 17 Jul 2007 19:15:10 -0500 http://www.voxforge.org/home/forums/message-boards/audio-discussions/you-can-now-submit-speech-to-voxforge-using-your-telephone Just go to this link: Submit Speech Using Your Telephone, http://www.voxforge.org/home/forums/message-boards/audio-discussions/you-can-now-submit-speech-to-voxforge-using-your-telephone Wed, 25 Apr 2007 14:00:02 -0500 http://www.voxforge.org/home/forums/message-boards/audio-discussions/errors-in-voxforge-corpus In the process of training Sphinx4, I'm finding there are some errors in the corpus. I've encountered one or more of the following errors: 1) Prompt doesn't match recording 2) Prompt has incorrect recording label 3) Prompt file named transcripts.txt 4) Prompt has a typo 5) Recording is unintelligible I'm wondering if and how I should report these findings and if they will be corrected in the repository. For #3 above, I'm wondering if there is some standard. In addition to the name of the prompts file, some prompts are all uppercase while some are mixed, some have recording labels pointing to the mfc directory while most are relative paths to the wav file. Some prompts have punctuation while some don't. Some prompts have multiple sentence fragments, while most are single sentences or a series of words. Thanks. http://www.voxforge.org/home/forums/message-boards/audio-discussions/errors-in-voxforge-corpus Thu, 12 Apr 2007 22:24:59 -0500 http://www.voxforge.org/home/forums/message-boards/audio-discussions/automatic-segmentation-of-librivox-audio email from David Gelbart: http://www.voxforge.org/home/forums/message-boards/audio-discussions/automatic-segmentation-of-librivox-audio Fri, 02 Mar 2007 15:05:21 -0600 http://www.voxforge.org/home/forums/message-boards/audio-discussions/mp3-podcast-audio-as-a-corpus-audio-source Email sent to Udhyakumar Nallasamy: Hi Udhyakumar, http://www.voxforge.org/home/forums/message-boards/audio-discussions/mp3-podcast-audio-as-a-corpus-audio-source Tue, 13 Feb 2007 14:14:56 -0600 http://www.voxforge.org/home/forums/message-boards/audio-discussions/more-on-collecting-speech-audio-for-free-gpl-speech-corpus My email to Joe Picone, ISIP (Institute for Signal and Information Processing) http://www.voxforge.org/home/forums/message-boards/audio-discussions/more-on-collecting-speech-audio-for-free-gpl-speech-corpus Tue, 13 Feb 2007 09:47:45 -0600 http://www.voxforge.org/home/forums/message-boards/audio-discussions/comments-on-a-good-acoustic-model-needs-to-be-trained-with-speech-recorded-in-the-environment-it-is-targeted-to-recognize Creating a new thread from comments made by David Gelbart in another thread: http://www.voxforge.org/home/forums/message-boards/audio-discussions/comments-on-a-good-acoustic-model-needs-to-be-trained-with-speech-recorded-in-the-environment-it-is-targeted-to-recognize Thu, 08 Feb 2007 21:06:51 -0600 http://www.voxforge.org/home/forums/message-boards/audio-discussions/-what-are-best-practices-for-collecting-speech-for-a-free-gpl-speech-corpus This is taken from a post I made to the comp.speech.research newsgroup: Hi, http://www.voxforge.org/home/forums/message-boards/audio-discussions/-what-are-best-practices-for-collecting-speech-for-a-free-gpl-speech-corpus Tue, 06 Feb 2007 12:49:27 -0600 http://www.voxforge.org/home/forums/message-boards/audio-discussions/brough-turner-on-creating-large-speech-corpora http://www.voxforge.org/home/forums/message-boards/audio-discussions/brough-turner-on-creating-large-speech-corpora Mon, 05 Feb 2007 20:50:00 -0600 http://www.voxforge.org/home/forums/message-boards/audio-discussions/issues-in-collecting-speech-audio-for-free-gpl-speech-corpus Email discussion I had with Arthur Chan (author of the article Do we have a true open source dictation machine?) Hi Arthur, http://www.voxforge.org/home/forums/message-boards/audio-discussions/issues-in-collecting-speech-audio-for-free-gpl-speech-corpus Mon, 29 Jan 2007 09:10:18 -0600 http://www.voxforge.org/home/forums/message-boards/audio-discussions/sample-freq-issue-not-covered-by-faq email from Robin: http://www.voxforge.org/home/forums/message-boards/audio-discussions/sample-freq-issue-not-covered-by-faq Thu, 21 Dec 2006 12:09:32 -0600 http://www.voxforge.org/home/forums/message-boards/audio-discussions/librivoxs-audacity-tutorial---how-to-clean-up-background-noise Here is a link to LibriVox's Audacity tutorial http://www.voxforge.org/home/forums/message-boards/audio-discussions/librivoxs-audacity-tutorial---how-to-clean-up-background-noise Sun, 12 Nov 2006 15:25:20 -0600 http://www.voxforge.org/home/forums/message-boards/audio-discussions/creating-a-cheap-quotrecording-studioquot Shortly put, creating a good recording place boils down to two things: 1) Eliminating external noise.2) Breaking up as much surface as possible to avoid echo.  As to the elimination of external noise, there is only so much you can do without spending a small [or huge] fortune: Pick a room that is the furthest away from trafic-noise. Close doors and windows. Shut the blinders/pull the curtains. (I take it that you have read the documentation, so telling you to turn off the aircondition/fan etc. should not be nessecary at this point). Now we get to the FUN part! You see, the art of braking surface is the art of doing what your mother told you never to do: Making A Mess(TM)! Thats right. What you need to do is to "scientifically" make a mess of the room. First, if there is no carpet on the floor, spreading out books with about a foot apart is a good start, but don't forget to make them stand up open if they can. Also, moving all the plants you have in the other rooms into your recording studio gives good results, as plants have a huge surface. Preferably the plants are placed on chairs, or the like, evenly distributed in the room. But the big problem is the walls... bare walls kill good recordings! Closets, "littered" shelves, racks and framed pictures help a lot here. Just remember that pictures with glass covers are actually worse than a bare wall, as glass bounces more sound than wallpaper! And while we are at it.. so does the hard unbroken surface of a door. The only easy/cheap way I can come up with is to place a mattress in front of it, or if it has a hook, hang your biggest coat on it. Then you systematically inspect the room to check if you can come up with a solution to every surface you see: Can you stand something in front of it?  Can you move it out of the room? Can you pull a blanket over it? Use poster-gum to fasten something to it? Etc. Be inventive!  Once your homebrew recording studio looks pretty much like a warzone you are ready to create clear and noise-free recordings... that is... if you can grab hold of a decent microphone! Have fun making a mess and recording :-) /macavity--FSF Associate member number 3423. http://www.voxforge.org/home/forums/message-boards/audio-discussions/creating-a-cheap-quotrecording-studioquot Thu, 12 Oct 2006 07:33:15 -0500