Voice datasets recorded for phonetic coverage
Structured, rights-clear recording sessions built for model training, not casual voiceover.
Conversational, natural delivery for commercial and corporate content — a versatile voice for brands, agencies and production companies.
Voice Lab research for production decisions
A practical research library on presence, intelligibility, AI voice identity and the way a recorded voice behaves in real production contexts.
Open Voice LabVoice Lab: AI Voice Identity & Conversational Presence
Research library on voice identity design for conversational AI, personality expression in text-to-speech systems and the way AI voice performs in user interaction contexts.
How an AI voice session is actually structured
Before recording, usage and consent are discussed directly — what the voice is for, how it will be used, and what rights apply.
Sessions are built for consistency across long blocks: the same tone, pace and mic distance from the first prompt to the last.
Files are delivered segmented and labeled for dataset processing or product integration, not as one continuous recording.
- Consent discussed before recording
- Consistent tone across long sessions
- Segmented files for dataset use
- Direction for conversational pacing
Where this fits
Structured, rights-clear recording sessions built for model training, not casual voiceover.
A TTS dataset needs something a normal voiceover session never worries about: complete phonetic and emotional coverage, recorded with total consistency across hundreds of lines. That changes the whole session structure, not just the script.
Coverage gets planned before recording — which sounds, which emotional ranges, which sentence structures need a take — rather than discovered gaps after the fact. Licensing and usage terms are settled before capture starts, since a dataset has a very different rights profile than a single ad read.
- structured voice dataset recording
- session discipline for phonetic and emotional coverage
- usage discussion before capture
Questions producers ask before booking
When is this the right service?
This service is useful for TTS and voice cloning projects where the recording must stay consistent, legally clear and useful for model training, with a professional Spanish voice recorded from Spain and prepared for international production standards.
What should be included in the brief?
Send the script, target market, usage, deadline, pronunciation notes, reference tone and whether the session needs live direction.
Can this support AI voice or dataset work?
Yes. AI voice projects can include consistent takes, consent-aware usage discussion, phonetic coverage and delivery prepared for technical review.
How does the process reduce revisions?
Tone, pace, rights, file format and audience are clarified before recording, so the first delivery is closer to the edit and campaign context.
Can JOIA record voice over projects from Spain for international clients?
Yes. Joel De Las Heras Bean records from a professional studio in Spain for agencies, brands, producers and technology teams working in Spain and internationally.
What languages are available?
Professional recordings are available in Spanish, English for premium and luxury brand contexts, and Catalan.
What AI voice work actually requires from a performer
Fatigue control matters more here than in a single ad read — hours of prompts need to sound like the same voice at minute five and minute two hundred.
Joel treats consent and usage as a conversation to have before recording starts, not paperwork to sort out afterward.
The performance has to work for a machine to learn from and for a human to listen to without noticing the difference.
- Stable tone across extended recording blocks
- Consent and usage discussed before capture
- Phonetic coverage for dataset completeness
- Direction for natural conversational turn-taking
Recording for systems, not just for ears
A dataset session and a conversational-assistant session ask for different things, so the direction is set before recording, not adjusted after the fact.
Text-to-speech and voice cloning work needs a consistent voice identity across hundreds of takes — the studio setup stays fixed for exactly that reason.
Where the voice is going matters: a customer-service assistant needs different pacing than a narrated onboarding flow, even from the same voice.
Choose the right next step
Still defining the voice identity for your product?
Listen to a demo close to a conversational or assistant use case.
Audio demoBuilding a dataset or TTS pipeline?
Send the prompt list or script sample, and get notes on coverage and session structure.
Script reviewReady to scope a recording session?
Share the intended use, volume and rights model so the proposal matches how the voice will actually be deployed.
Availability and quoteWhat AI teams get
Consent discussed upfront, consistent tone across long sessions, and files segmented for technical review.
Contact