comeagain?

Inspiration

Communication is the most important skill in the world, I believe technology should enable this, not take away from our ability to do so. After filming my first reel last week, I realized how important it is to be able to play back our own voices, because the clarity of communication completely changes how others are perceived. I've always been perceived as pretty quiet, even if my personality is outgoing, because my voice tends to stay in my throat versus projecting outwards. If I had this years ago, I would've made so many more friends and been so much more successful.

What it does

World's first tone eval system that allows you to receive frontier insights instantly about your speech. Eight speech dimensions being measured with a hand-rolled DSP pipeline in numpy- batched autocorrelation for pitch tracking (bias-corrected, parabolic-interpolated, octave-error filtered), RMS/spectral-band analysis for volume and articulation, and a de Jong & Wempe–style intensity-peak detector for syllables/pacing- plus faster-whisper (CTranslate2, base.en, int8, CPU-only) for local transcription that makes intonation and stress judgeable against sentence form. Scores are computed against literature-derived target bands with soft shoulders (85–100 in-band, tapering to 25, floor below), and coaching is either a deterministic rule engine or, when an ANTHROPIC_API_KEY is set, a byLLM call that phrases the same measured numbers without ever inventing one.

How we built it

Jac, claude code, and a lot of credits

Challenges we ran into

Hosting it on jachammer wasn't working because the speech processing components (llvmlite, faster-whisper, and cmudict) have no pre-built wheels for the Jac runtime

Accomplishments that we're proud of

Instead of relying on generic AI feedback, we engineered a pipeline that measures delivery across multiple dimensions like pacing, pitch, articulation, stress, and intonation before generating coaching. We're especially proud that every piece of feedback is grounded in measurable signal processing, making the advice objective instead of subjective. We also redesigned the experience to feel like a personal coach rather than a technical analytics dashboard.

What we learned

This project taught us that building AI products isn't just about using large language models. The quality of the output depends heavily on the quality of the inputs. We learned a lot about digital signal processing, speech analysis, and designing meaningful metrics that people can actually improve upon. We also gained experience integrating traditional DSP techniques with LLMs to create feedback that is both technically accurate and easy to understand.

What's next for tone

We plan to release this publicly on producthunt, gather users, run it by actual Ted Talks and debate coaches, and ultimately make the world a more understandable place.

Built With

• jac