Talkie

Speech library for Arduino

Quick start

Download the .zip file from the link at the top of the page. Install the "talkie" directory in your Arduino / libraries directory. Restart the Arduino software, and pick one of the examples from the Talkie section.

Connect headphones or an audio amplifier to digital pin 3 on Arduino, pin 9 and 10 on Leonardo, Flora and Lilypad USB, or the DAC pin on Teensy.

Note that currently, Talkie works on 168 or 328 based Arduinos at 16MHz and with reduced clarity at 8MHz. It works on Teensy LC, 3.1 & 3.2. It is being tested on 32u4 devices by replacing the Timer2 PWM with Timer 4. Teensy Leonardo Flora Lilypad USB

TODO: update TimerInterrupt with support for ARM Zero devices 12-bit fidelity on Teensy 100bit fidelity on ARM Zero Increase sample rate on ARM?

Introduction

It is a software implementation of the Texas Instruments speech synthesis architecture (Linear Predictive Coding) from the late 1970s / early 1980s, as used on several popular applications:

• Texas Instruments Speak & Spell family of educational products
• Texas Instruments TI-99/4A Speech System expansion
• Acorn BBC Micro Speech Synthesiser expansion
• Atari arcade games (eg. Star Wars series, Indiana Jones, Gauntlet)
• Apple ][ Echo 2
• IBM PS/2 Speech Adapter

Talkie comes with over 1000 words of speech data that can be included in your projects. Most words only take a fraction of a KB, so you can add plenty.

How to use the library

See the examples. Any commented data lines in the examples can be un-commented and used. Note that some speech libraries come with word endings such as '-S', '-Z', '-TEEN' that can be used to increase vocabulary.

Word dictionaries

• Vocab_UK_Acorn - a male UK english voice. 165 words related to home computing. 16K bytes in total. Data originally part of Acorn Computers Speech Synthesiser, and famously voiced by BBC's Kenneth Kendall.

• Vocab_US_Clock - a female US english voice. 35 words related to time of day. 4K bytes in total. Data of unknown origin from ROM serial number VM71003A.

• Vocab_US_TI99 - a deep male US english voice. 360 words related to home computing. 32K bytes in total. Data originally part of Texas Instruments TI-99/4A Speech System.

• Vocab_US_Male - a male US english voice. 206 words related to aeronautics. 16K bytes in total. Data of unknown origin. ROM serial number VM61002.

• Vocab_US_Male_Large - a male US english voice. 402 words related to aeronautics. 48K bytes speech data. Data of unknown origin. ROM serial number VM61003/4/5.

• Demo_Toms_Diner - demo of the work-in-progress Talkie compressor. 24K bytes music data. Tom's Diner was written/recorded by Suzanne Vega (Amazon UK) (Amazon US) or look for 'Solitude Standing' on your preferred music store. (Only this album contains the original a capella version)

The Audio output

Talkie sets up a special very high speed PWM, so audio can be taken directly from pin 3 with no other filtering. On Teensy, a true analog output DAC is used.

Note that Talkie uses Timers 1 and 2 for PWM & timing, which may conflict with PWM outputs or other libraries. On Teensy, one of the available IntervalTimers is used.

Future plans

1. I'm working on an encoder for generating your own recordings (See demo). Its not producing results of the quality I would like, but things are improving. For now Qboxpro, an unsupported old Windows application, can produce Talkie compatible data streams.
2. Improve compatibility with other Arduinos and AVR devices
3. Implement rule based infinite vocabulary synthesis from Computer Concepts Speech ROM and Texas Instruments Terminal Emulator 2.

Important facts

• LPC currently outputs speech sounds on digital pin 3 only. This is currently hard-wired within the library.
• Tivoice uses Timer 1 and Timer 2, which can break analogWrite(), and some other libraries.

Thanks

• For motivation and encouragement: Linx Electronics Club, esp. Rob, Shaw, Steve
• Thanks to the original Texas Instruments speech group: Larry Brantingham, Paul Breedlove, Richard Wiggins and Gene Frantz
• Thanks to the emulator-heads for their extensive documentation and coefficient tables: Lord Nightmare, Kevtris and all MAME and MESS contributors
• Thanks to the authors of low bandwidth open source speech codecs for their inspirational source code: Asterisk, Codec2, Hawkvoice, Speex
• Scene greets: LFT, Mahoney

This library was developed using Arduino and Freemat.