mirror of
https://github.com/storytold/artcraft.git
synced 2026-11-03 00:47:53 +00:00
- Vowels indicate syllable - Algorithm tracks consonants and uses heuristics to attach them to a vowel island. - Still work in progress.
2.2 KiB
2.2 KiB
Remaining Tasks
HIGH LEVEL GOALS
- Sound better
- Handle load
- Recoup costs
- Look good (UI, speaking capability) - good reflection
- Good code
Audio Samples
- MORE SAMPLES
- IMPROVED SAMPLES
- Remove extra space between words (using parsing to insert spacing)
- No more missing phonemes
- Real versions of ['AO0', 'AW0', 'AW2', 'AY0', 'EY0', 'OY0', 'UH0'], which were backfilled with fakes.
Tokenizing
- Plurals
- Times (9 p.m. ET)
Parsing
- (Everything in tokenizing)
- Decimal numbers
- Collapse pauses, eg: "Foo,,," -> "Foo (breath)"
- Numbers that are years (2016)!
- Twitter hashtags, usernames - conjoined word parsing (difficult)
Synthesis
- Get rid of static with "Uhhh", "Ummm" of variable length (option).
- Syllable System:
Cleanup
- Refactor into a very well-defined "SpeakRequest" with all parameters, including new ones like "is_twitter", "missing_word_mode = slience, record_pop, umm", etc.
- Remove old query parameters
- Set default volume
Scaling
- Load test
- Read wave files into memory
- Multi-server, with nginx as load balancer
- Prevent others from calling it
- Caching strategy
Server
- Favicon, .ico and apple icon
- robots.txt
Hardening
- Reduce API surface
- Limit length of input sentences
- Rate limiting
Presentation
- Better frontend UI
- Testing page (current frontend)
Monetization
- Can I get sued?
- Ads
- Native Apps
- (I don't have time to learn this.)
- Contact other developers?
- PhoneGap? (Blah)
Extra
- Clientside speed changing
- Let people type nonsense syllables
Parsing Examples
- Camel case even outside of hashtags:
- "Google acquires FameBit", StateOfW, AdaLovelaceDay
- Time: 7:30pm, 5:30ET, 9pm ET, 8:00 P.M., 8pm ET
- Hyphenated: 2-party
- They will get horrible sentences, violently.We have the support.#RepealThe1st
- #100YearsStrong