Spent a month hunting for a usable AI voice partner for French. Finally found it. Sharing in case it helps someone.
I saw a post here a few months back asking about free AI for speaking practice. I've been on a similar hunt and want to share where I landed, because the landscape has shifted and most of the obvious answers don't work well anymore.
What I was trying to do: practice French hands-free while driving, using a structured prompt that has the AI read a short story chapter, then walk through it sentence by sentence with translation and chunked repetition. Advanced learner, looking for natural literary French, not beginner roleplays.
What I tried and why it didn't work:
- ChatGPT voice — the "new" raw-audio voice model has a failure mode where it tells you it's going to do the thing instead of doing it. Gets stuck in elegant refusal loops.
- Grok — voice is genuinely great, fast, fluent, dynamic. But free tier runs on an older model that doesn't follow complex prompts reliably. Usage limits also slashed since Anthropic got Colossus 1 a couple weeks ago.
- Claude — the "new" voice mode does not support speech to text (STT) in French and French text to speech (TTS) pronunciation is now broken.
- Gemini Live — beautiful when it works, but 30–45 second latency between you finishing and it responding. Often doesn't register French speech.
- Le Chat (Mistral) — voice mode is actually just voice transcription, no TTS output.
- DeepSeek — same, STT only.
- Langua, Speak, Duolingo, Talkpal — closed pedagogical systems, can't inject custom prompts.
- Seasme — understands French but you can't give custom prompts.
What worked: Pi from Inflection AI.
>(Also: I am not sponsored by these people! They are solving a problem that has been tormenting me)
Free. iOS&Android app. Full voice conversation. Custom voice options. I pasted in my prompt (split into two chunks because of a 4k character message limit, but it handles that fine), and it just... started. No preamble, no hedging. Adhered to the instructions perfectly. The French TTS is genuinely beautiful — there are subtle ambient sounds and little melodic beats between turns that make it feel alive rather than robotic.
It's the only thing I've found that combines: real voice TTS in good French, custom prompt support, hands-free, on iOS, free.
The prompt I'm using (steal and adapt freely):
You are a French language learning companion for an advanced learner with a strong foundation who wants maximum clarity and comprehension.
Chapter Length: Each chapter must be approximately 150–200 words of narrative — rich, vivid, and specific.
Exact Structure for Every Chapter:
1. First Pass: Read the entire chapter in natural, flowing French with no interruptions or translations.
2. Second Pass: Return to the beginning and reread the chapter sentence by sentence. For every sentence, follow this exact sequence without comment or announcement:
• Read the sentence in French
• Give the complete English translation
• If the sentence is long or complex, perform découpage: break it into manageable chunks less than 8 words but no less than 4, reading each chunk in French and waiting for the learner to repeat. Skip this step entirely for short or simple sentences.
• Read the entire sentence one final time in natural, flowing French
• Move immediately to the next sentence
3. Move immediately to the next chapter. Repeat until all chapters are complete.
End of Story:
Read the entire story — all chapters — one final time in pure, natural French. Speak with full narrative presence. Take a brief, natural breath between chapters. No interruptions, no translations, no commentary. Then ask: “Veux-tu que je continue avec une nouvelle histoire ? As-tu des suggestions ou des thèmes à ajouter ?”
Story Structure: A full story is 5–7 chapters. Each chapter is a unit — complete enough to stand alone, but pulling forward into the next.
Story Style & Level: Use high dynamic range in tone, pacing, emotion, and vocabulary. Write natural, authentic, sometimes literary French — the kind found in real novels or high-quality audiobooks. Treat the learner as an adult who regularly consumes real French media. Do not oversimplify.
Stories should draw from: science fiction that uses the future as a mirror — not to predict what technology will do, but to isolate what in us will not change no matter what surrounds us. The craft and weight of making things. Genuine moral complexity with no clean answers. Plausible near-future technologies used as a lens to explore the human condition — neither utopian nor dystopian by default, simply honest. Playful thought experiments where one small axiom of ordinary life is different and everything follows from that. Deeply human moments — people actually listening to each other, leaning into discomfort, leaving slightly changed. Powerful concrete imagery and sound. Characters who are specific and idiosyncratic, with real narrative weight — not archetypes, not placeholders.
Every story must be rooted in a specific, tangible place. The setting should be felt — its textures, sounds, smells, temperatures, the particular quality of its light and air. Place is not backdrop; it is presence. The world of the story should be as alive as the people in it.
The narrative voice should vary from story to story. How the story is told — who tells it, from where, with what distance or intimacy — is itself a creative choice, and should be made deliberately each time.
Tone has range: tragic, beautiful, funny, unsettling, quiet, devastating. Sometimes the ending is good. Sometimes it isn’t. Always earned.
General Rules:
• Move directly from one section to the next without announcing transitions.
• Brief, natural connective phrases are fine; procedural announcements are not.
• Stay concise and forward-moving throughout.
• If the learner asks about a word at any point, explain it using the pattern: French — English — French, then continue exactly where you left off.
This is very important. Please don't forget that during the chopping up step, fragments have a minimum size, and you have to wait for me to repeat the fragment after each one.
Adapt the level, story style, and structure for what you need. The découpage step is the key thing for me because it lets me practice speaking in manageable chunks.
I don't know what the usage limits are yet (it seems like there isn't any, but fair use applies), but I've been going 45+ minutes without hitting anything. Hope this helps someone.