Project 2: Speech enhancement

In this project, algorithms for speech enhancement are designed, implemented (in Matlab), evaluated, and documented. The physical outcome of the project is a technical report motivating the algorithm design and describing the implementation and evaluation of the algorithms.

Assignment

• Design and build a multi-microphone speech enhancement/beamforming system for far-end noise reduction.
• Use matlab
• Generate signals according to the signal model discussed in class using the audio files and impulse responses given below.
• Perform an evaluation of the speech enhancement system.

Data sets

Speech signals:

• Clean speech 1
• Clean speech 2
• Babble noise
• Aritificial nonstationary noise
• Stationary speech shaped noise

Multimicrophone impulse reponses:

• Impulse responses

The matfile contains 5 sets of impulse reponses. Each set contains the impulse responses from a source location to the 4 microphone locations. Each of these 5 sets can then be convolved with a sound signal to model the signal at each of the micrphone locations.

Literature

To evaluate the speech quality of your algorithms, you can use the algorithms described in the following literature:

• An Evaluation of Intrusive Instrumental Intelligibility Metrics - Steven Van Kuyk et.al.
• An Instrumental Intelligibility Metric Based on Information Theory - Steven Van Kuyk et.al.
• Speech Quality Assessment - Grancharov & kleijn
• STOI - Cees Taal

Consultants

dr.ir. R.C. (Richard) Hendriks

Tel: +31 15 27 83544
Email: R.C.Hendriks@tudelft.nl
Room: HB 17.050