Skip to content
Turing College Logo
Submit projectSign In

AVA: Voice and communication Coach

AVA: Voice and communication Coach

About this project

An AI coach for helping you on how you speak, not only what you say: vocal variety, filler words, pitch, clear answers and staying calm in difficult conversations. Practice first, then let it review your real conversations.

Goal and problem

People often form an impression of us before they fully understand what we’re saying. The way we speak- our tone, pace, confidence, clarity, and ability to get to the point— can make a big difference in how our ideas are received. Yet most of us rarely get honest, constructive feedback on how we communicate. We don’t hear ourselves the way others do, and people around us may not feel comfortable pointing out what we could improve.

Ava is a Voice and Communication Coach to give that feedback: measured, specific, and private.

1. Practice. Read a famous speech aloud and compare your pace, pauses and vocal variety with the original speaker: where they slowed down, paused, or lifted their voice. Practice interview questions and difficult situations out loud, such as a manager pushing back or a tense conversation at home. After each answer you see a strong and a weak reply to the same question. Every word's pronunciation is checked, and you can hear any word and repeat it until it's right.

2. Real conversations. Upload a recorded interview, meeting or presentation. The coach separates the speakers, recognises your voice and reviews only you:

- your filler words and hedging;

- how often you start with your point;

- how you handle pressure;

- how your vocal variety holds up.

You get scores on a five-part communication rubric, each with evidence you can replay, and you can ask follow-up questions like "How often did I say 'kind of'?"

Key features

- Vocal variety: pitch range, pace and pauses, compared with great speakers

- Filler words: um, uh, "kind of" and hedging, counted and shown in context

- Difficult situations: practise interviews and tough conversations out loud, with model answers

- Pronunciation: word-by-word check, listen-and-repeat

- Real conversation review: your voice only, rubric scores with evidence, replay at any moment

- A coach with memory and a follow-up chat that looks up your own words

- Private by design: audio and recordings never leave your laptop

Tech stack

- Backend: Python 3.11, FastAPI, LangGraph , OpenRouter

- Speech: faster-whisper, torchaudio MMS_FA forced alignment, wav2vec2 phoneme recognition, espeak-ng, pyannote speaker diarization, Kokoro text-to-speech, parselmouth and librosa

- Data: PostgreSQL 17 (built in), SQLAlchemy, Alembic

- Observability and testing: LangSmith tracing, pytest, LLM evaluation scripts

- Frontend: plain HTML, CSS and JavaScript

Created bymaryam Ramezani-Bartsch
Published atNovember 19, 2026
CourseAI engineering
Open LinkedIn