Open Source Japanese Conversation Practice With AI (Local Speech, One API Key)

Описание: I got tired of paying $15-18/month for AI conversation practice apps that don't even let you pick the model. So I built my own.Tama (Talk-based Adaptive Mastery Assistant) is a free, open-source desktop app for practicing Japanese conversation. It uses local speech-to-text (Whisper) and local text-to-speech (Voicevox) - the only thing you pay for is the LLM, and you get to choose which one.I've been using Mistral 3 through Open Router and haven't even hit 1 cent of usage yet.

GitHub: https://github.com/rodolfo-terriquez/...
Voicevox: https://voicevox.hiroshiba.jp/
Open Router: https://openrouter.ai/

What's inside:
12 conversation scenarios (convenience store, restaurant, doctor, hotel, and more)
Ongoing conversations with AI personas that remember you
Custom scenario builder
JLPT N5→N1 adaptive difficulty
30+ Japanese voices (Voicevox)
Post-session grammar corrections and vocabulary feedback
Built-in SRS flashcards — vocabulary from your conversations feeds back into future practice
Text mode and voice mode
Full conversation history with grammar breakdowns

How it works:
Voicevox handles text-to-speech (runs locally, bundled with the app)
Whisper handles speech-to-text (runs locally on your machine)
You just need one API key for the LLM (Open Router recommended — access any model)

0:00 Introduction - Why I Built This Project
0:10 The Challenge of Learning Japanese Conversation
1:07 Current Options: Tutors, Language Exchange, AI Apps
2:22 Problems with Existing Apps (Duolingo, Pingo, Tenmin)
7:04 Introducing Tama - Talk-Based Adaptive Mastery Assistant
7:24 Why Tama is Better: Open Source, All-in-One, Ridiculously Cheap
9:21 Demo: Setting Up Tama
9:41 Downloading and Installing the Application
10:26 First Look at the Interface
10:59 Scenarios - Practicing Common Daily Situations
12:06 Conversations - Long-Form Dialogue Practice
14:28 Flashcard System with Spaced Repetition
15:48 History - Reviewing Past Conversations
17:24 Settings Overview
17:44 Choosing Your Japanese Level
18:03 Selecting Response Length
18:31 Large Language Model Provider Setup
19:07 Text-to-Speech Engine (Voicevox)
20:40 Voicevox Project Overview
21:34 Speech Recognition Options
22:46 Starting a Conversation Demo
24:58 Grammar Feedback and Vocabulary Review
26:42 Cost Breakdown - Under 1 Cent for Multiple Sessions
28:17 What's Next for Tama
29:29 Multi-Language Support Possibilities
29:49 Closing Thoughts