Tag Archives: Speech synthesis

Looking at Hidden Markov Models (HMMs)

While I think the synthesis from synSinger is intelligible, it’s certainly not very natural. The technology behind synSinger is quite old-school: rule-based formant synthesis. The rest of the world has since moved on to using HMMs – Hidden Markov Models. … Continue reading

Posted in Uncategorized | Tagged , , | Leave a comment

Added nasal resonator and anti-resonator

Per Klatt, I’ve finally added a nasal resonator and anti-resonator. The resonator is the same two-pole resonator as in MITalk, which has a two-sample history: constant MINUS_PI_T = -math:PI / audio:SAMPLE_RATE constant TWO_PI_T = 2 * math:PI / audio:SAMPLE_RATE — … Continue reading

Posted in Uncategorized | Tagged , | 1 Comment

Day of the Locus

The two current goals of synSinger are clarity and naturalness. The current output is wide of both marks, although it’s slowly getting better. The initial values of for the phonemes were taken from published tables from the MITalk program, although … Continue reading

Posted in Uncategorized | Tagged , , | Leave a comment

Talk It!

In searching for other examples of formant synthesizers, I ran across TalkAny, aka “Talk It!”. It’s an repackaging of SoftVoice’s  formant synthesizer.  The reviewer notes negatively that it is: Old-fashioned and out of date Voices not natural However, after listening … Continue reading

Posted in Development | Tagged , , , | Leave a comment

Aligning PItches and Notes

One issue I’ve been having is with “dropout”, where certain sounds would simply disappear. I finally tracked it down to a mis-match between the pitch curve and the phoneme list. Phonemes are aligned so that the vowels fall at the … Continue reading

Posted in Development | Tagged , , | Leave a comment

Working with Silence

The “warbling” sound was happening because it was interpolating between one large formant target and another. In particular, it was interpolating into what should have been silent targets, with a value of 0. I’ve code in place to check for … Continue reading

Posted in Development | Tagged , , , , | Leave a comment

Some Progress

Enough of the code is in place to render the single line “Twinkle, twinkle little star.” Or, more accurately, /T W IH NG K AH L  T W IH NG K AH L  L IH T AH L  S T … Continue reading

Posted in Development | Tagged , , , , | Leave a comment