http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/using-auxiliary-sources-of--knowledge-with--htk please, I need you help me.I seek how USING AUXILIARY SOURCES OFKNOWLEDGE such as the pitch frequency, the short-termenergy and ate-of-Speech in HTK tanks http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/using-auxiliary-sources-of--knowledge-with--htk Sun, 04 May 2008 10:53:07 -0500 http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/why-headapt-doesnt-work I'm trying to follow the tutorial to adapt my acoustic model but it doesn't work. I've installed HTK 3.2.1, but the command HEAdapt still is not known by the system. What can I do? Any help? thanks!! http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/why-headapt-doesnt-work Sun, 06 Apr 2008 20:34:05 -0500 http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/htk-and-sphinx-training-tutorials Today I came across something by Keith Vertanen, who is involved in the speech version of Dasher (Speech Dasher): http://www.inference.phy.cam.ac.uk/kv227/papers/baseline_wsj_recipes.pdf It is a detailed explanation of how he set up HTK & Sphinx training for the Wall Street Journal corpus, with URLs that can be used to download the training scripts. He wrote, "My goal is to provide practical advice and results to researchers who are thinking of using HTK or Sphinx for real-time recognition on dictation-like tasks". While WSJ is a proprietary corpus, his work could still be useful as a source of examples. His scripts have support for monophones, word-internal triphones, and cross-word triphones. Also, he says "Many of the acoustic models used in the experiments described later in this paper are available for download" He also has some language model training scripts at http://www.inference.phy.cam.ac.uk/kv227/ http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/htk-and-sphinx-training-tutorials Thu, 03 Apr 2008 15:12:15 -0500 http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/managing-single-character-words I'm running into an error with julius after compiling my model apparently without an error. The problem seems to be related to my attempt to use single character words apart from A and I. First choke is with 'B'. I have tried the BEEP and the voxforge lexicons and get the same error. I had to add many single character words to the BEEP lex. Anyone see what I am missing? Is there some special precaution when using letters of the alphabet? Here's what julius gives me: STAT: *** loading LM00 _default STAT: reading [sample.dfa] and [sample.dict]... Error: voca_load_htkdict: line 16: triphone "*-b+iy" or biphone "b+iy" not found Error: voca_load_htkdict: line 16: triphone "b-iy+*" or biphone "b-iy" not found Error: voca_load_htkdict: the line content was: 14 [B] b iy http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/managing-single-character-words Fri, 21 Mar 2008 17:27:43 -0500 http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/mfcc-format-of-julius-4_0 Dear Voxforge developers: I encountered a problem that when i use the MFCC_0_D_A_Z as the feature parameter type,the recognition performance of the mic live input is degraded compared to that using the MFCC_0_D_N_Z ? But the performance of mfcc file input which using MFCC_0_D_A_Z is greater than that which using MFCC_0_D_N_Z. Why? Is it that any other Configuration option must be set ? http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/mfcc-format-of-julius-4_0 Mon, 10 Mar 2008 08:33:09 -0500 http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/accessing-howto In a number of posts I see references and links to a howto which puts much of the tutorial into scripts. However, the link http://www.voxforge.org/home/dev/howto just leads back to a welcome page that has no reference to a howto. Am I missing something obvious? http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/accessing-howto Tue, 04 Mar 2008 06:18:44 -0600 http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/error-reading-speech-data-from-wav-files I'm testing my acoustic modes uder Linux and using the wav files used for training.But julius seems can not read speech data from most of these files(it decodes 20 out of 100 utterances,but the other 80 are failed).The error infos are shown as follows: ### read waveform input Error: gzfile: unable to open wav/1.wav Error: adin_file: failed to open wav/1.wav Error: adin_file: failed to read speech data: "wav/1.wav " ...... the wav files are recorded under Windows by Audacity.Can't the files recorded under Windows be decoded correctly under Linux? Does anybody know the reason?thanks http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/error-reading-speech-data-from-wav-files Tue, 04 Mar 2008 01:06:34 -0600 http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/error-in-test-acoustic-model-using-julius When testing the acoustic model I ran the instructions: julian -input rawfile -filelist wavlst -smpFreq 48000 -C julian.jconf > juliusOutput (copied from "step2-creating test prompts" ) but did not get the correct results.The error infomations were: -------------------------------- ###### check configurations ###### initialize input device ###### build up system Reading in HMM definition...(ascii)...limit check passed defined HMMs: 50 logical names: 506 in HMMList base phones: 44 used in logical done Making pseudo bi/mono-phone for IW-triphone...369 added as logical...done Reading in dictionary... line 18: triphone "*-z+ih" or biphone "z+ih" not found line 18: triphone "z-ih+r" not found line 18: triphone "ih-r+ow" not found > 6 [ZERO] z ih r ow error in reading sample.dict: 1 words failed out of 18 words ERROR: failed to read dictionary, terminated Terminated ------------------------------ i compare my tiedlist with the one given in "Step 10 - Making Tied-State Triphones",and they are completely the same,and files sample.grammar and sample.dict are also copied from "step2-creating test prompts". That is why? http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/error-in-test-acoustic-model-using-julius Sat, 01 Mar 2008 06:36:48 -0600 http://www.voxforge.org/home/forums/message-boards/speech-recognition-in-the-news/where-is-the-step-11 Hello, I would like to ask where is the step 11 from the HTKbook. For me who as a very bad recognition, I think it is important to know if it comes from the hmm or not !! When I write the first command of the step 11, HVIte he tells me that I have a problem of Target. Can you explain the difference between MFFC_0 and MFCC_D_N_Z_0 ??? An over question is why the promp in the tutorial doesn't always correspond to the grammar? Thanks http://www.voxforge.org/home/forums/message-boards/speech-recognition-in-the-news/where-is-the-step-11 Fri, 29 Feb 2008 09:11:40 -0600 http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/help-for-sphinx-spanish-acoustic-model-to-htk-julius-acoustic-model-format Hello, I need to convert these sphinx acoustic model spanish files to Julius HTK: http://www.speech.cs.cmu.edu/sphinx/models/hub4spanish_itesm/ I am a newbie on this, could you please help me? I am trying to use the Speech http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/help-for-sphinx-spanish-acoustic-model-to-htk-julius-acoustic-model-format Fri, 29 Feb 2008 05:11:36 -0600 http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/merging-accoustic-models__ I was wondering if it is possible to merge a completed accoustic model (let's say CMU Communicator, because it's the best 'free' one that I know of at the moment) with your own training data, or even another accoustic model (say the one from VoxForge for example)? Or would I have to scrap the CMU communicator and start from scratch... Currently I am getting about a 70% recognition rate with CMU communicator using Sphinx 2 on a large vocabulary over VoIP. I would hate to have to scrap my 'success' and start from scratch with a potentially a much higher WER. Speaking of which, what kind of success rates have people been seeing on an AM trained with the VoxForge data? http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/merging-accoustic-models__ Sun, 24 Feb 2008 09:01:13 -0600 http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/step-2-error I am trying to create acoustic model using auto script but I am getting very strange error in step 2. It says something like error in HDMan, cat: ./interim_files/monophones1, no such file or directory and then in the next line something like cat:logs/Step2_HDMan.log no such file or directory It looks like there is some problem with cat command but I can`t figure out what and how it is connected to HDMan. And why there is no monophones1 file when HDMan is supposed to create it. I have 10 words (non english) in my dictionary and I added them all to voxforge lexicon at the begining of file. http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/step-2-error Tue, 19 Feb 2008 11:42:58 -0600 http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/step-10-error I need to make grammar for julian that recognizes names of few languages as the native speakers say (english, deutsch, espanol etc.) I need to build my own acoustic model of course. So first I wanted to see if I can train him to recognize just one language (I choose Deutsch). I followed the steps in tutorial (automated, with script) and because it says I need more words with same phones, I added more words that contain phones as Deutch does. Those are chow, dutch, toy, annoy. And I recorded 2 wav files containing Deutsch Chow Dutch, and the other Toy Annoy. So I started the script (as it says in tutorial) and everything goes well until step 10. Then it says: making hmm13 ERROR [+2662] AddUnseenCommand: there are no existing trees FATAL ERROR - Terminating program C:\cygwin\HTK\htk-3.3-windows-binary\htk\HHEd.exe making hmm14 ERROR [+5010] InitSource: Cannot open source file ./interim_files/tiedlist ERROR [+7010] InitHMMSet: Can't open list file ./interim_files/tiedlist ERROR [+2321] Initialise: MakeHMMSet failed FATAL ERROR - Terminating program C:\cygwin\HTK\htk-3.3-windows-binary\htk\HERest.exe making hmm15 ERROR [+5010] InitSource: Cannot open source file ./interim_files/tiedlist ERROR [+7010] InitHMMSet: Can't open list file ./interim_files/tiedlist ERROR [+2321] Initialise: MakeHMMSet failed FATAL ERROR - Terminating program C:\cygwin\HTK\htk-3.3-windows-binary\htk\HERest.exe cp: cannot stat `./interim_files/hmm15/hmmdefs': No such file or directory cp: cannot stat `./interim_files/tiedlist': No such file or directory What could be the problem? Thank you! http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/step-10-error Wed, 30 Jan 2008 19:37:05 -0600 http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/hhed-error-2662 Hi, I am running a Chinese(Mandarin) LVCSR, but I came across the error below, while I am running the below clusterring command: HHEd -A -D -V -B -w clustering/hmm10_1000_800/MMF -d http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/hhed-error-2662 Fri, 18 Jan 2008 02:16:10 -0600 http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/need-help-in-supporting-indian-languages Hi, I want to add support for Indian languages and publish the Language Model. Can somebody guide/advise me to start with. Regards sanjit http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/need-help-in-supporting-indian-languages Sun, 06 Jan 2008 07:20:00 -0600 http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/hvite-log How to understand HVite_log file numbers. What is meaning of Ac, LM, Act. For example in my log file they are: [1176 frames] -48.0691 [Ac=-56529.3 LM=0.0] (Act=4.9). It is good or bad result. What is good score1 in Julian Recognition output. Is it better when score is lower? http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/hvite-log Tue, 11 Dec 2007 14:22:32 -0600 http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/disagreement-in-tutorial Why in tutorial http://www.voxforge.org/home/dev/acousticmodels/windows/create/htkjulius/tutorial/monophones/step-7 Step 7 - Fixing the Silence Models: change matrix in <TRANSP> to 3 by 3 array change numbers in matrix as follows: numbers are: 0.0 1.0 0.0 0.0 0.9 0.1 0.0 0.0 0.0 but in example numbers are: <TRANSP> 3 0.0 0.7 0.3 0.0 0.6 0.4 0.0 0.0 0.0 <ENDHMM> http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/disagreement-in-tutorial Sun, 02 Dec 2007 06:49:38 -0600 http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/voxforge-acoustic-model-parameters Email from David Gelbart: Hi Ken, Congratulations on your 1 year anniversary and reaching 27 hours of data. I'm writing to comment on a technical issue that I'm not sure if you are aware of: The general rule I have seen with ASR systems is that, as the amount of training data increases, it eventually becomes necessary to add more acoustic model parameters in order to get the full benefit of the additional data. On the other hand, using too many acoustic model parameters may cause overfitting (in other words, the system starts modeling quirks of the training data to the point where the system's performance on non-training data is worsened). Thus, you may need to periodically tune the number of acoustic model parameters you are using. I suppose the easiest way to do this is to create a test set which does not overlap with the training set, and measure word recognition accuracy on the test set for various acoustic model sizes. One way to increase the number of parameters is to use more Gaussians in the Gaussian mixtures. (One way to do this in HTK is to add one or more additional mixup stages. This has the advantage that you can use your test set to compare recognition accuracy before and after the mixup, so that you can obtain your recognition accuracy numbers without having to retrain a system from scratch each time.) Another way to increase the number of parameters is to move from monophones to triphones (unless you are using triphones already). Another way is to reduce the amount of state-tying. Regards, David http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/voxforge-acoustic-model-parameters Wed, 17 Oct 2007 20:28:24 -0500 http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/error-when-compiling-model---newmacro-macro-or-model-name-st_o_2_1-already-exists Hi! I tried to compile a model using your howto-approach and ran into a small (big) problem when "making hmm13". "ERROR [+7063] NewMacro: macro or model name ST_O_2_1 already exists FATAL ERROR - Terminating program C:\cygwin\HTK\htk\HHEd.exe" I tried a fulltextsearch for the name and it returned no matching file. According to the HTK-manual the error code simply means that HHEd tries to assign a name that is already used. Any ideas? Help would be really appreciated! Bettina (team simon) http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/error-when-compiling-model---newmacro-macro-or-model-name-st_o_2_1-already-exists Thu, 09 Aug 2007 01:43:10 -0500 http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/mfcc-parameters-and-scripts Dear developers... Is it possible to calculate just the mfcc, energy and deta energy parameters? When i use MFCC_E_D, HTK calculates delta for both MFCC and energy and I just want to calculate for the energy. I didn´t find out this information on HTK book. Is it possible to create a script file to automate the training of my HMMs? Using tutorial i have to type all command lines. Thanks for your help Marcos http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/mfcc-parameters-and-scripts Mon, 30 Jul 2007 14:39:01 -0500 http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/mfcc_d_n_z_0-format Dear Voxforge developers, I am asking about details of the MFCC format used in your acoustic models. The global settings in hmmdefs (Julius_AcousticModels_16kHz-16bit_MFCC_O_D_build726.tgz) is as follows: ~o<STREAMINFO> 1 25<VECSIZE> 25<NULLD><MFCC_D_N_Z_0><DIAGC> In step 5 of the data preparation tutorial you suggest to use TARGETKIND = MFCC_0_D How do MFCC_D_N_Z_0 and MFCC_O_D agree? And what is the exact meaning (and order) of the 25 vector components in hmmdefs? For these qualifiers: _D Delta coefficients appended_N Absolute log energy suppressed_Z Cepstral mean subtracted_0 Cepstral C0 coefficient appended I would expect (12+1)*2=26 components (NUMCEPS=12, + C0, doubled by delta coefficients), not 25 components. Thanks Martijn http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/mfcc_d_n_z_0-format Sat, 21 Jul 2007 03:24:40 -0500 http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/hdman-character-encoding-problem Hi folks, Let us assume my lexicon file content is like: ÄHNLICH ae n l i C and my wordlist file content is: ÄHNLICH Now when I run HDMan command like following : HDMan -m -w wordlist -n monophones1 -l dlog dict lexicon everything works fine except 'dict' file. dict file contains following line: \303\204HNLICH ae n l i C sp As you can see, 'Ä' character replaced with some ascii numbers. I tried to set my LANG, LC_ALL environment variables and other LC=XXX variable's value to de_DE.ISO8859-1 ( on FreeBSD machine ) but no difference. Could somebody tell me what I am doing wrong? Any help would be appreciated. Nagi http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/hdman-character-encoding-problem Wed, 06 Jun 2007 03:49:58 -0500 http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/htk-error-in-step-10 hello there it might be related to what was discussed here, so i'm posting here i'm getting an error in step 10 like this Step 10 - Making Tied-State Triphones ============================================================== making hmm13 ERROR [+2662] AssignStructure: cannot find tree for b state 2 FATAL ERROR - Terminating program C:\cygwin\HTK\htk-3.3-windows-binary\htk\HHEd.exe I also submitted this error on VoxForge Dev Any help would be appreciated lucian http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/htk-error-in-step-10 Wed, 16 May 2007 21:50:33 -0500 http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/speech-recognition-model-converter http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/speech-recognition-model-converter Thu, 03 May 2007 10:48:59 -0500 http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/using-the-models-as-source-for-speech-synthesis2 As we are building models to be used as HMM for the recognizer, is there a way to use them to actually do TTS ? I'm a bit new to the subject so apologies if the question seems stupid. thanks! http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/using-the-models-as-source-for-speech-synthesis2 Wed, 04 Apr 2007 14:52:08 -0500 http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/using-the-models-as-source-for-speech-synthesis As we are building models to be used as HMM for the recognizer, is there a way to use them to actually do TTS ? I'm a bit new to the subject so apologies if the question seems stupid. thanks! http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/using-the-models-as-source-for-speech-synthesis Wed, 04 Apr 2007 13:01:13 -0500 http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/humans-are-great-but-why-not-use-commercial-and-oss-text-to-speech-engines-too Hello all, I'm new here. Lately I've been absorbing everything I can get my hands on regarding speech recognition for a few projects I have in mind. I've read Arthur Chan's letter to the community and I agree with his assessment. In addition, I'm hoping this site will be instrumental in bringing the community it's first open source text/speech corpus. I spent more than a few hours playing with Sphinx in it's various incarnations tonight and well ... it's a start. I'll likely be downloading from this site soon in an attempt to make my own acoustic model. (Right after I get a decent mic) But anyway, on to the reason for this message: Why don't we auto generate a couple hundred hours of text to speech from festival and any cheap commercial text-to-speech engine out there that we can get our hands on? Sure, it won't be as good as a human read corpus, but with all the different voices and accents (i.e. british vs. american, and male vs female) it's got to be better than nothing, right? Besides, is bad speech really undesirable? Don't we want less than perfect samples? Won't that make the model more robust? What do you think? Cheap couple hundred hours? http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/humans-are-great-but-why-not-use-commercial-and-oss-text-to-speech-engines-too Sat, 31 Mar 2007 02:11:20 -0500 http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/acoustic-model-creation-error-in-step-10 I'm building a new voice grammar and repeately I get an error in step 10 making hmm 13. I have tried both the script-way and the hard oldschool tutorial;-) I have even tied just to run it with the original data and voice but the same error-message accur every time: ERROR [+2662] FindProtoModel: no proto for p in hSet FATAL ERROR - Terminating program HHEd making hmm14 ERROR [+5010] InitSource: Cannot open source file ./interim_files/tiedlist ERROR [+7010] InitHMMSet: Can't open list file ./interim_files/tiedlist ERROR [+2321] Initialise: MakeHMMSet failed FATAL ERROR - Terminating program HERest making hmm15 ERROR [+5010] InitSource: Cannot open source file ./interim_files/tiedlist ERROR [+7010] InitHMMSet: Can't open list file ./interim_files/tiedlist ERROR [+2321] Initialise: MakeHMMSet failed FATAL ERROR - Terminating program HERest cp: cannot stat `./interim_files/hmm15/hmmdefs': No such file or directory cp: cannot stat `./interim_files/tiedlist': No such file or directory Additional I thought of replacing the old julius-3.5.2 with the new julius-3.5.3. There should not be any problems just doing this or is there..? http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/acoustic-model-creation-error-in-step-10 Thu, 01 Mar 2007 10:04:30 -0600 http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/single-gram-representation_ I am pretty new to speech recognition, but I don't clearly see why single gram phones cannot cover pretty much all words. Why do we need N-grams? Is it simply for word separation? Thanks! http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/single-gram-representation_ Wed, 21 Feb 2007 11:07:24 -0600 http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/one-word-grammar-always-recognized Hi everybody, http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/one-word-grammar-always-recognized Thu, 25 Jan 2007 11:55:39 -0600 http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/librivox-data Someone on Slashdot pointed out that LibriVox data might be useful for you.  The full text of public domain books is often available online, which could serve as a transcript.  Segmentation of the audio into smaller chunks might be needed for training. (I'm not sure if it would be; I'm used to training with speech files separated into sentences but I don't know if that's necessary.) But if so, maybe an automated forced alignment against the text could be used to do that.  http://www.voxforge.org/home/forums/message-boards/acoustic-model-discussions/librivox-data Thu, 12 Oct 2006 13:28:04 -0500