DOC PREVIEW
Automatic Volume Leveler for Real Time Speech Applications

This preview shows page 1-2-3 out of 10 pages.

Save
View full document
View full document
Premium Document
Do you want full access? Go Premium and unlock all 10 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 10 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 10 pages.
Access to all documents
Download any document
Ad free experience
Premium Document
Do you want full access? Go Premium and unlock all 10 pages.
Access to all documents
Download any document
Ad free experience

Unformatted text preview:

BIEN 2011Automatic Volume Leveler for Real Time Speech ApplicationsJustin BareDr. Carol Espy-WilsonDr. Tarun PruthiBIEN 2011The Volume LevelerVolume LeveledNoisy Noise ReducedNoise ReductionVolume LevelingBIEN 2011Motivation• Noise reduction is useful• But it has problems:– Attenuation of speech– Changes in color or timbre and loudness0 0.5 1 1.5 2 2.5 3x 104-1-0.500.51Time (in samples)AmplitudeNoisy Speech and Noise-Reduced Speech Noisy SpeechNoise-Reduced SpeechNoiseSpeechBIEN 2011Overview• My goals for the automatic volume leveler– Fix speech attenuation– Restore loudness– Ensure no clipping occurs in output signal– Do all of the above under real-time constraints• Only present and past frames of signal are available• For each frame (0.01 seconds) of signal:Input noisy speech frameNoise removal (e.g. spectral subtraction)Automatic volume levelerOutput frameBIEN 2011AlgorithmInput noise-reduced signal frameVoice Activity Detector (VAD)Frame has no speechUpdate noise trackingDo not amplifyFrame has speechEstimate the Signal-to-Noise Ratio (SNR)High SNR: Amplify to original levelMiddle SNR: Amplify conservativelyLow SNR: Do not amplifyPrevent clippingBIEN 2011Ideal vs. Real VAD• Ideal VAD (not obtained in real time)• Real VAD (obtained in real time)0 0.5 1 1.5 2 2.5 3x 104-0.8-0.6-0.4-0.200.20.40.60.81Clean SignalPhonetic TranscriptionTimeSound1.78‘f’1.9‘aa’2.02‘r’0 0.5 1 1.5 2 2.5 3x 104-0.200.20.40.60.81VAD Decision0 0.5 1 1.5 2 2.5 3x 104-1-0.8-0.6-0.4-0.200.20.40.60.81Noisy SignalVAD Algorithm0 0.5 1 1.5 2 2.5 3x 104-0.200.20.40.60.81VAD DecisionBIEN 2011Ideal VAD vs. Real VAD0 0.5 1 1.5 2 2.5 3x 104-1-0.500.51Time (in samples)AmplitudeVolume-Leveler Using Ideal VAD NoisyVolume-LeveledNoise-ReducedIdeal VAD0 0.5 1 1.5 2 2.5 3x 104-1-0.500.51Time (in samples)AmplitudeVolume-Leveler Using Real VAD NoisyVolume-LeveledNoise-ReducedReal VADBIEN 2011Ideal VAD vs. Real VAD: SNR and Noise Level Increase-12dB-3dB 0dB 3dB 6dB 12dB18dB01234567Using Ideal VADSignal-to-Noise Ratio (dB)Amount of Increase (dB) SNR IncreaseTotal NoiseLevel IncreaseNear-SpeechNoise Level Increase-12dB-3dB 0dB 3dB 6dB 12dB18dB01234567Using Real VADSignal-to-Noise Ratio (dB)Amount of Increase (dB)BIEN 2011Volume Restoration-12dB -3dB 0dB 3dB 6dB 12dB 18dB020406080100120Percent of OriginalSpeech Amplitude AttainedSignal-to-Noise Ratio (dB)Level of Restoration ofSpeech to Original Amplitude (%) Volume-Leveled Speech (Ideal VAD)Volume-Leveled Speech (Real VAD)Noise-Reduced SpeechBIEN 2011Future Work• Less reliance on VAD accuracy• Incorporate coloring/timbre restoration• Implement in a fast, low level language such as CAcknowledgments• National Science Foundation OCI award #1063035• Dr. Carol Espy-Wilson• Dr. Tarun Pruthi• MERIT BIEN faculty and


Automatic Volume Leveler for Real Time Speech Applications

Download Automatic Volume Leveler for Real Time Speech Applications
Our administrator received your request to download this document. We will send you the file to your email shortly.
Loading Unlocking...
Login

Join to view Automatic Volume Leveler for Real Time Speech Applications and access 3M+ class-specific study document.

or
We will never post anything without your permission.
Don't have an account?
Sign Up

Join to view Automatic Volume Leveler for Real Time Speech Applications 2 2 and access 3M+ class-specific study document.

or

By creating an account you agree to our Privacy Policy and Terms Of Use

Already a member?