How to make my BBC MicroBit talk using MicroPython
With the demonstration of general unit selection synthesis inEnglish in Rob Donovan's PhD word and ATR's CHATRsystem ( and ) by the end of the 90's unitselection became the hot topic for speech synthesis research. Howeverinspite of very high quality examples of it working, generalized unitselection also produces some very bad quality synthesis. As theoptimial search and selection agorithms used are not 100% reliableboth high and low quality synthesis is produced.
Acorn BBC Micro Peripherals - Computing History
Into the 2000s a new statistical method of speech synthesis has cometo the forefront. Again pioneered by work on Japan. Prof KeiichiTokuda's HTS System (from Nagoya Institute of Technology) showed thatbuilding generative models of speech, rather than selecting unitinstances can generate reliable high quality speech. Its prominancecame to the fore front at the first Blizzard Challange in 2005 whichshowed that HTS output was reliably understoof by listeners. HTS, andso-called HMM synthesis seems to do well on smaller amounts of data,and when then the data is less reliably recorded which offers asignificant advantage over the requirement of very large carefullylabelled corpora that seem to be required for unit selection work. Weinclude detailed walkthroughs form CMU's CLUSTERGEN statisticalparametric synthesizer which is tightly coupled with this Festvoxvoice building toolkit, though HTS continues to benefit from theFestival systems (and much of what is in this document).
Of course with faster machines and large disk space people began to lookto improving synthesis by using larger, and more varied inventories forconcatenative speech. Yoshinori Sagisaka at ATR in Japan developednuu-talk in the late 80s early 90s, which used a muchlarger inventories of concatenative units thus instead of one example ofeach diphone unit there could be many and a automatic acoustic basedselection was used to find the best selection of sub-word units from afairly geenral database of speech. This work was done in Japanese whichhas a much simpler phonetic structure than English making it possible toget high quality with still a relatively small databases. Though evenby 94 the generation of the parameter files for a new voice in nuu-talk(503 senetences) would take several days of CPU time, and synthesiswas not generally real-time.
Acorn BBC Micro Speech Synthesiser expansion;
The 8-bit 6502A microprocessor can directly address 64kilobytes of memory. In the BBC Microcomputer, 32 kilobytes ofthis memory is contained within Read Only Memories (ROM's) . ROMis non volatile; that is, it retains data when the power to it isremoved. The Machine Operating System (MOS) of the microcomputeris contained within 15 kilobytes of this read-only memory on a 16kilobyte device, (normally IC51). The machines address decodingcircuitry "maps-out" the remaining 1 kilobyte, sincethis area is devoted to JIM, FRED and SHEILA. The BASICinterpreter is contained within another mask-programmed, 16kilobyte ROM (normally IC52).
Let's make some noise! It's the BBC micro:bit MusicFest
However, although we are now at the stage were talking computers arewith us, there is still a great deal of work to be done. We can nowbuild synthesizers of (probably) any language that can producereconizable speech, with a sufficient amount of work; but if we are touse speech to receive information as easily when we're talking withcomputers as we do in everyday conversation, synthesized speech mustbe natural, controllable and efficient (both in rendering and in thebuilding of new voices).
BBC Computer Related Manuals: Advertisements
Techniques were being developed to compress speech in a way that itcould be more easily used in applications. The Texas InstramentsSpeak 'n Spell toy, released in the late 70s, was one of the earlierexamples of mass production of speech synthesis. The quality was prettypoor, but for the time it was very impressive. Speech was basicallyencoded using LPC (linear Predictive Coding) and mostly used isolatedwords and letters though there were also a few phrases formed byconcatenation. Simple TTS engines based on specialised chips becamepopular on home computers such as the BBC Micro in the UK and theApple ][.
Speech Synthesis and Cartridge ROM system for the BBC Microcomputer:
A feature of the BBC Microcomputer is its ability to perform a"hardware scroll". The start-address-register withinthe 6845 CRTC is used to refer to the start address of screenmemory. By altering the contents of this register it is possibleto scroll the "active" screen display area. In the BBCMicrocomputer, the hardware scroll is implemented in a moresophisticated fashion.