1 / 10

Speaker Recognition Experiment

Speaker Recognition Experiment. Seungchan Lee Intelligent Electronic Systems Human and Systems Engineering Department of Electrical and Computer Engineering. Overview. Software Release lm_tester, network_builder Debugging utility : Purify HierarchicalSearch Class

cameo
Télécharger la présentation

Speaker Recognition Experiment

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Speaker Recognition Experiment Seungchan Lee Intelligent Electronic Systems Human and Systems Engineering Department of Electrical and Computer Engineering

  2. Overview • Software Release • lm_tester, network_builder • Debugging utility : Purify • HierarchicalSearch Class • Speaker Recognition System • Do Experiment with HMM using the MFCCs • Adding invariants in the feature file • Next Plans

  3. Software Release • Purify • Resolve Compilation problem • Compile without sphere utility • Track down memory problem • Resolve HierarchicalSearch class problem • Namemap problem • This is caused by the difference of the system which extract features and train the feature files.

  4. Baseline system set up • Set up Baseline system • It takes much time to set up the baseline system at this time. • It will be faster at the next experiment. • Problems with Adding invariants • Transform_builder • Window Class • Lyapunov exponent • The result recipe sof file has some problems.  Every result feature vector has the same value.  Sundar helps me to add the Lyapunov exponent, we did not find the reason.

  5. Baseline system set up • Set up Baseline system • It takes much time to set up the baseline system at this time. • It will be faster at the next experiment. • Problems with Adding invariants • Transform_builder • Window Class • Lyapunov exponent • The result recipe sof file has some problems.  Every result feature vector has the same value.  Sundar helps me to add the Lyapunov exponent, we did not find the reason. @ Sof v1.0 @ @ FeatureFile 0 @ name = "FEATURES"; file_type = "TEXT"; file_format = "SOF"; ………………………….. values = { 11.502,11.4683,12.1384,10.292,5.05849,8.20064,3.72829,8.71881,0.655161,4.4856 ,1.29329,-1.04087,-24.0325 }, { 11.502,11.4683,12.1384,10.292,5.05849,8.20064,3.72829,8.71881,0.655161,4.4856 ,1.29329,-1.04087,-24.0325 }, { 11.502,11.4683,12.1384,10.292,5.05849,8.20064,3.72829,8.71881,0.655161,4.4856 ,1.29329,-1.04087,-24.0325 }, ………………………………

  6. Baseline system set up • Transform_builder problem • Open existing recipe file and save as different recipe file. • Two recipe files are not the same. • Window class has been modified after saving • Ryan looked into this problem alignment = LEFT; normalization = NONE; compute_mode = CROSS_FRAME; duration = 0.025; debug_level = NONE; constants = LEFT; alignment = NONE; normalization = CROSS_FRAME; compute_mode = 0.025; duration = NONE;

  7. Experiment Result • Problems with hypothesis • The result hypothesis scores is highly less than previous result • The “ACCEPTED” and “REJECTED” hypothesis does not give correct decision. MFCC Feature Test Utterance Test Utterance IHD, 16 Mixture model JSGF, 16 Mixture model ACCEPTED: 1.70586e-05 (KAAA) REJECTED: -0.000698179 (KAAB) ACCEPTED: 2.19704e-05 (KAAC) REJECTED: -0.00218273 (KAAD) REJECTED: -0.000529905 (KAAG) ACCEPTED: 0.000170417 (KAAH) REJECTED: -0.000673878 (KAAJ) REJECTED: -0.842606 (KAAA) REJECTED: -2.53688 (KAAB) REJECTED: -0.827423 (KAAC) REJECTED: -0.803932 (KAAD) REJECTED: -1.15936 (KAAG) REJECTED: -1.00537 (KAAH) REJECTED: -0.555092 (KAAJ)

  8. DET Curve DET Curve (Tang’s) Probability Plot

  9. Next Plan • Software Release • Help Daniel to finish the release. • Speaker Recognition System • Do Experiment with HMM using the MFCCs and invariants • Compare the result to previous result • Analyze the result and track down the problem • Nonlinear System • Find the research topics

  10. List of activities (Spring semester) • Become familiar with our system • Make first simple program • Software Release • Work with Daniel on the Software Release • ProductionRuleTokenType– modify read/write function with namemap class • lm_tester, network_builder • Dummy symbol, exclude symbol problem • Debugging utility : Purify • Resolve compilation problem ( without sphere utility) • Fix memory problem – segmentation fault (HierarchicalSearch class • Speaker Verification System • Designing interface for Verification System • Combine three functionality, HMM/GMM, SVM, RVM  New isip_verify Utility • Resolve checksum error (namemap)

More Related