Speech Synthesis Methods, Implementation and A Coprocessor with Leon

1
Speech Synthesis - Methods, Implementation and
A Coprocessor with Leon

• Rajeev Bakshi
• M Tech Comp Sc and Engg

2
Outline

• Speech synthesis technologies and vocal system
• Klatts speech synthesizer
• Architecture design approach
• CEERIs approach
• FPGA design flow
• Methodology to attach a coprocessor at the fpu
  interface of Leon processor

3
TTS Text To Speech synthesizer

• Automatic speech signal generation on computers
  is commonly called speech synthesis.
• Input can be in many of the forms. It can be
  written text, optical characters, parameters, etc

4
Applications

• Telecommunication Services
• Language Education
• Aid to handicapped persons
• Talking books and toys
• Vocal Monitoring
• Multimedia, man-machine communication.
• Fundamental and Applied Research

5
Three Broad Classes

• Speech synthesis technologies can be broadly
  categorized into three categories
• Concatenate synthesis
• Formant synthesis
• Articulatory synthesis

6
Vocal System
7
Concatenate Speech Synthesis

• Waveform segments are stored in a database
• For a given text, these segments are joined based
  on some joining rules
• Audible glitches because of transition between
  segments
• Efficient lookup and searching is necessary to
  locate the segments

8
Formant Speech Synthesis

• Formants Resonant frequencies that occur at the
  main resonant areas of the vocal tract for a
  given sound
• It consists of artificial reconstruction of the
  formant characteristics to be produced.
• This is done by exciting a set of resonators by a
  voicing source or noise generator to achieve the
  desired speech spectrum
• The addition of a set of anti-resonators
  furthermore allows the simulation of nasal tract
  effects, fricatives and plosives.

9
Articulatory Speech Synthesis

• Articulators are Speech organs
• It is based on computational models of the human
  vocal tract and the articulation processes
  occurring there.
• Most natural sounding speech synthesis.
• Attempt to describe the actual speech production
  mechanism.

10
Comparison of Speech Synthesis Techniques
11
Outline

• Speech synthesis technologies and vocal system
• Klatts speech synthesizer
• Architecture design approach
• CEERIs approach
• FPGA design flow
• Methodology to attach a coprocessor at the fpu
  interface of Leon processor

12
Klatts Model

• Cascade/Parallel formant synthesizer
• Source filter model
• Voicing source
• impulsive model
• Turbulent noise
• random number generator
• Vocal tract transfer function
• resonator combination

13
Klatts Speech Synthesizer
14
Overall System
words
.par
.raw
rules
klatt
input text
essentially a lookup table
parameter file
cascade parallel formant synthesizer
waveforms ascii file/ binary file
Private to an organization (ceeri) Language
dependent (hindi, ..)
Available in public domain Language independent
15
Source Filter Model klatt
Source
White Noise
Vocal Tract Filter
speech
Impulses (f0)
16
Source Filter Model

• Production of Speech
• Generation of Sound Source at the glottis at some
  point along the length of the vocal tract
• Filtering of these sources by the vocal tract
• Sound Sources
• Quasi Periodic Sources
• Turbulence Noise Sources

17
Source Filter Model

• Turbulence Noise
• Turbulence occurs due to rapid air flow at a
  constriction. Turbulence noise can be produced at
  a constriction at the glottis, or at a
  constriction made with the tongue or lips above
  the glottis.
• Aspiration Noise is produced at a glottal
  constriction e.g. h
• Frication Noise is produced at a supraglottal
  constriction e.g. s, f, v
• Noise is modeled by the pseudo random number
  generator

18
Source Filter Model
19
Resonator band pass filter Antiresonator band
stop filter
20
Digital Resonator
O/P Seq y(nT)
I/P Seq x(nT)
y(nT) Ax(nT) By(nT-T) Cy(nT-2T)
Where ninteger
21
Digital Resonator

• Input output characteristics of resonator are
  specified by
• Resonant Frequency (Formant), F
• Resonance Bandwidth, BW
• C -exp(-2piBWT)
• B 2exp(-piBWT) Cos(2piFT)
• A 1- B - C

22
Outline

• Speech synthesis technologies and vocal system
• Klatts speech synthesizer
• Architecture design approach
• CEERIs approach
• FPGA design flow
• Methodology to attach a coprocessor at the fpu
  interface of Leon processor

23
Architecture Design Approach

• Criteria
• Performance, Space Optimization, Semantic Gap,
  Cost, etc.
• ASICs
• Disadvantages
• Higher cost
• No flexibility
• DSPs
• Disadvantages
• Higher cost

24
Architecture Design Approach

• CISCs
• Disadvantages
• Relatively Slower
• More Power Consuming
• RISCs
• Disadvantages
• Large semantic gap with the application
• ASIPs
• Advantages
• Reduced semantic gap
• Reduced power consumption

25
Outline

• Speech synthesis technologies and vocal system
• Klatts speech synthesizer
• Architecture design approach
• CEERIs approach
• FPGA design flow
• Methodology to attach a coprocessor at the fpu
  interface of Leon processor

26
CEERIs work

• ASIP for Hindi TTS Voice Chip
• Rule chip and voice chip (Sw and FPGA)
• Rule Chip (SW)
• Database lookup and concatenation rules to
  generate parameters for a given sentence
• Voice Chip (HW)
• Klatts model of vocal cord which takes
  parameters and produces amplitude samples

27
Outline

• Speech synthesis technologies and vocal system
• Klatts speech synthesizer
• Architecture design approach
• CEERIs approach
• FPGA design flow
• Methodology to attach a coprocessor at the fpu
  interface of Leon processor

28
FPGA Design Approach

• HDL compilation
• Logic Optimization
• Technology Mapping
• Placement
• Routing
• Static timing analysis
• FPGA configuration file generation

29
FPGA Design Flow 1
30
FPGA Design Flow - 2
31
Outline

• Speech synthesis technologies and vocal system
• Klatts speech synthesizer
• Architecture design approach
• CEERIs approach
• FPGA design flow
• Methodology to attach a coprocessor at the fpu
  interface of Leon processor

32
Methodology

• Leon core provides two interfaces
• Coprocessor interface
• FPU interface
• FPU interface
• fpu_core
• meiko_fpu

33
Methodology
34
Thanks

• Project Home Page -
• http//karnali.cse.iitd.ernet.in/srijan/text2speec
  h/