A personal intention.
Three ways to hear it.
Shift Audio turns everyday intentions into a personalized song, a guided relaxation track and clearly audible affirmations. Claude connects the user's preferences to three consistent, original texts. ElevenLabs gives the relaxation and affirmations their spoken voice.
Listen to the complete French demo →Start with three short excerpts · How the demo was made
Complete French pack generated: four first-party Claude API calls produce the coordinated texts; three audio tracks have been generated and delivered for a fictional French profile. English and bilingual text tests are complete; audio review is ongoing.
The experience
Personalized song
Original lyrics and music shaped by the user's intention, musical style, mood and energy.
Guided relaxation
Gentle, optional suggestions for a quiet pause, followed by a return to everyday attention.
Audible affirmations
Clear speech at natural speed over music, with the full text available to read. No hidden messages.
Claude is central to the product
- Analyze the intention. Claude checks the request against the general wellbeing scope and reframes suitable goals around the user's own choices and actions.
- Write three coordinated texts. Three Claude calls create song lyrics, a relaxation script and affirmations using a shared context.
- Respect preferences. Language, emotions, grammatical agreement and included or excluded expressions shape the texts. Missing, truncated or contradictory outputs stop before audio production.
- Create and deliver audio. Suno through KIE.ai creates music; ElevenLabs produces speech; FFmpeg mixes the tracks. Cloudflare R2 hosts MP3 files and the listening page.
Current integration: claude-sonnet-4-6, called directly through the first-party Anthropic Messages API from n8n. Claude writes the text; it does not generate the audio waveforms. API credits would support this existing integration.
Claude writes. ElevenLabs speaks.
Coordinated texts
Claude creates the relaxation script and affirmations from a shared intention and the user’s preferences.
Personalized speech
ElevenLabs Text to Speech, using eleven_v3, voices both tracks in the selected language and voice. Grammatical agreement is a separate preference.
A clear listening experience
We assemble the spoken segments and mix them over a Suno instrumental background, keeping the voice at natural speed and in the foreground.
ElevenLabs is an existing product dependency. Its offer in the Claude Startup Stack would support pronunciation, pacing and multilingual voice evaluations, subject to the partner’s eligibility and redemption terms.
A focused initial use case
Example: “I want to speak up in meetings, one step at a time.” The product can combine acoustic pop, a calm voice and action-focused phrases. It does not claim that colleagues will react in a specific way or that a promotion is guaranteed.
The initial audience is adults seeking a personal music and relaxation routine. The product is intended as general wellbeing and creative entertainment, with no medical, neurological, financial or relationship outcome promised.
About the demonstrations
The three excerpts use a fictional profile, Alex, and an ordinary confidence goal: speaking up in meetings. Claude generated the texts. Suno created the song, singing and instrumental background; ElevenLabs voiced the two spoken tracks.
Listen to 45 seconds of acoustic pop, 75 seconds of guided relaxation and 45 seconds of audible affirmations. A complete French pack has also been generated, mixed and uploaded to R2, with a listening page and all 28 affirmations. The song excerpt comes from an earlier run of the same fictional profile; the two voice excerpts come from the complete pack.
French, English and mixed-language texts have been generated and checked. In mixed mode the song is bilingual; relaxation and affirmations remain in French. English and mixed-language audio have not yet been validated.
Examples illustrate the product and are not customer testimonials or evidence of therapeutic effectiveness. The public demo record lists the source generations, providers, excerpt durations and remaining checks. Listening quality remains part of our prototype review.
Built with Claude
Alongside the working API integration, Claude has helped develop the product. At repository baseline 8154f04, 11 commits explicitly credit Claude Opus 4.6 as a co-author, out of 50 commits in that history. These markers document specific contributions; they do not measure every use of AI.
Examples include harmonizing the site and delivery page design, and improving the mobile FAQ. The public audio demonstrations show the current product's output.
Stage and next steps
French generation and audio delivery validated; commercial launch in preparation. Built by Ryan using AI-assisted development. A complete demonstration is available to listen to. The next steps are listening reviews, multilingual audio evaluation and connecting the updated generator to the commercial journey.
The planned starting offer is a one-time personalized pack, with an indicative price of €47.
Claude API credits would fund multilingual evaluations and early product tests on this existing integration. Claude Team would support the founder’s product planning, content review, documentation and development.
Boundaries
Requests involving treatment, weight-loss targets, coercion, self-harm or reinforcement of delusions are outside the product scope. Sensitive requests are stopped during analysis. These checks are model-based and require continued evaluation; they are not a guarantee of perfect detection.
Users can stop at any time, choose which suggestions to keep and read every affirmation. Relaxation should only be played while stationary in a safe setting, never while driving.