pesto Pi is a Greek text-to-speech synthesizer that runs on the $25 Raspberry Pi 700 MHz ARM computer. It is based on pesto TTS and its current version features the Greek small-footprint voice "Simos_1081".
pesto Pi runs smoothly on Raspberry Pi, performing over 60 x Realtime (*) on a Model B based prototype. Its size can vary depending on the compression and the phoneme database (current demo occupies 4.3MB and can be further compressed if required) starting from as low as 1MB of space in the SD card, while it uses around 100KB of RAM during runtime.
(*) "60 x Realtime" means that the TTS needs 1/60 of the duration of the synthesized speech to actually synthesize the speech. So, to synthesize a prompt of 10 seconds, the processing time is 167 milliseconds. Pretty fast on a 700MHz CPU, isn't it?
pesto Pi can run on both Models A and B.
C and Python libraries are available to include pesto Pi in your projects.
Voice "Simos_1081" is based on recorded speech diphone segments. Prosody has been modelled using machine learning techniques on natural speech. The small-footprint nature of this voice generates a mechanic hue, which suits applications like games, robots, alarms etc.
Voice "Simos_1081" supports the Greek language, while it can speak English with Greek accent.
Greek language processor features:
- Letter to phoneme converter for Greek: it can speak correctly any Greek word, even mispelled or weird letter combinations (unique feature), with no lexicon limitations.
- Light version of PhonAesthesia Language Suite context-based text normalizer for numeric formats (numbers, phone numbers, addresses, dates, time etc), abbreviations and acronyms.
English language processor features:
- Letter to phoneme converter for English: any pronouncable word is read as-is, otherwise it is spelled out (feature borrowed from flite). In any case, Greek phonemes are always used.
- Numbers are always read in Greek.
pesto Pi is currently available upon request. You may contact e-rhetor to find out more.