Formant estimation in Singing Studiovoicestudies/artts/doc/presentations/... · •Singing analysis...

1

Formant estimation in Singing Studio

PROJECT MEETING

Vítor Almeida

Faculdade de Engenharia da Universidade do Porto, Porto

January 19th, 2013

2

Formant estimation algorithm

Project Meeting January 19th, 2013

Features:

- Analysis window : 1024 samples, 75% overlap, sine window.

- Sampling frequency: 22050 Hz.

3

Formant estimation algorithm

Project Meeting January 19th, 2013

Estimation of noise component:

4

Formant estimation algorithm

Project Meeting January 19th, 2013

Estimation of formant candidates using LPC and Cepstrum:

5

Formant estimation algorithm

Project Meeting January 19th, 2013

Some difficulties:

• The existence of vibrato makes hard the noise estimation

• The closeness between two or more formants

• Finding a rule for the selection and validation the candidates

6

Formant estimation algorithm

Project Meeting January 19th, 2013

In SingingStudio:

7

Vibrato analysis in Singing Studio

PROJECT MEETING

Ricardo Sousa

Faculdade de Engenharia da Universidade do Porto, Porto

January 19th, 2013

8

Introduction

• Context and Objectives

• Singing analysis and bio-feedback applied to singing training/teaching

• Acoustic vibrato parameterization

Robust

Descriptive

Objective

Physiological meaning

• Automatic vibrato detection

• Visual interface of vibrato analysis

Project Meeting January 19th, 2013

9

Method

•Analysis Algorithm

Pitch Segment

selection

Vibrato Segment

detection

Parameter

Computation

Pitch Estimation

Voice

Signal

Pitch curve

Pitch

Segments

Vibrato

Segments

Vibrato

Parameters

Step 1

Step 2

Step 3

Step 4

January 19th, 2013 Project Meeting

-Spectral based methods

-Frame based

- FFT interpolation (accurate

F0 estimation)

10

Method

11

Method

•Vibrato percentage: Ratio of vibrato duration and the entire theme duration

•Mean Duration: Mean duration of all segments.

•Mean frequency: Mean frequency of all segments.

•Mean extension: Mean extension of all segments.

•Sinusoidal Purity: Measure the regularity of vibrato (similarity to sinusoidal waveform)

•Vibrato parameters

Duration

1/Frequency

Extension

January 19th, 2013 Project Meeting

12

Results

• Automatic detection of vibrato segments

Vibrato detection 1 Vibrato detection 2

January 19th, 2013 Project Meeting

Additional features:

• Manual adjustment of vibrato segments

13

Results

• Qualitative and Quantitative Evaluation

Irregular vibrato Regular vibrato

Observation:

• Qualitative evaluation: “Good vibrato”, “Bad vibrato”

January 19th, 2013 Project Meeting

14

Conclusion

• Qualitative and Quantitative evaluation of vibrato.

• Visualization and bio-feedback.

• Automatic analysis.

• Physiological interpretation.

• Interactive interface.

January 19th, 2013 Project Meeting

15

Truly grateful!

January 19th, 2013 Project Meeting