Chrisa — English Emotional Speech Dataset
8,158 clips · 3.50 hours · 6 emotions · 48 kHz, 32-bit float. €149 per dataset.
BedVibe Emotional Speech Datasets for AI Voice Training
Professionally recorded emotional speech corpora for TTS, voice cloning, expressive synthesis, and research.
Each voice includes 59–60 vocal effects such as moans, gasps, whimpers, screams, laughs, panting, breathing, inhales, exhales, crying and more.
Commercially Licensed Studio Data — Signed voice-actor agreements — Proof available on request
Every speaker in the published seven-speaker English corpus recorded the same 1,360-phrase manuscript across six emotional states (Neutral, Angry, Happy, Scared, Shouting and Whisper), for approximately 8,160 clips per speaker.
This shared phrase set ensures perfect alignment across all seven speakers — the same sentences, the same emotional coverage, and consistent structure throughout.
The corpus includes 1,344 structured phrases with questions, dramatic lines, pangrams, digits, written numbers, alphabet coverage, plus 16 long monologues.
Current datasets are available in English and Greek, and we are currently recording Norwegian. Additional languages such as French, Danish, Swedish, and other Scandinavian languages can be produced on demand. Contact us at bedvibe@bedvibe.studio for custom language datasets.
All recordings are professionally produced in studio environments. BedVibe uses owned, contracted, or internally recorded voice data for commercial production. Signed agreements and consent documentation are maintained privately for formal partner review where applicable. Contact us for licensing and commercial use details.
Click a portrait to purchase a dataset. Use the emotion buttons below each speaker card to preview real samples.
Elena — English Emotional Speech Dataset
8,158 clips · 4.56 hours · 6 emotions · 48 kHz, 32-bit float. €149 per dataset.
Gianna — English Emotional Speech Dataset
8,157 clips · 4.59 hours · 6 emotions · 48 kHz, 32-bit float. €149 per dataset.
Giouli — English Emotional Speech Dataset
8,158 clips · 4.50 hours · 6 emotions · 48 kHz, 32-bit float. €149 per dataset.
Synovia — English Emotional Speech Dataset
8,160 clips · 4.08 hours · 6 emotions · 48 kHz, 32-bit float. €149 per dataset.
Fotis — English Emotional Speech Dataset
8,159 clips · 4.02 hours · 6 emotions · 48 kHz, 32-bit float. €149 per dataset.
Pan — English Emotional Speech Dataset
8,160 clips · 5.44 hours · 6 emotions · 48 kHz, 32-bit float. €149 per dataset.
Full English Pack — All 7 Speakers Bundle
57,110 clips · 30.69 hours · 6 emotions · 48 kHz, 32-bit float. €599 bundle.
Custom Dataset Production
Custom multilingual dataset production for TTS, voice cloning, expressive synthesis, and research. Additional speakers, emotions, languages, or scripted content recorded to specification. Custom quote — contact bedvibe@bedvibe.studio.
Every dataset package also includes 59–60 bonus vocal effects per speaker: gasps, moans, sneezes, yawns, screams, crying, laughing, panting, breathing, whimpers, and pain and effort vocalizations.
Commercial License — Product & API
Full English Pack plus the right to distribute trained models, embed them in SDK, CLI or MCP software you ship, and run hosted inference for your own customers. €1,799 one-time price.
Enterprise License — OEM & Sublicensing
Everything in the commercial license plus the right to sublicense trained models to your own customers, white-label and OEM distribution, performer-rights indemnity, and speaker-disjoint expansion. €4,999 one-time price.
Greek Voice Datasets
Original studio recordings created before October 2026: the 1,360-phrase script recorded in Greek in six emotional states. Commercial AI training, speech synthesis and dataset licensing rights are available subject to the applicable BedVibe licence terms. Pan is Panagiotis Gkilis, founder of BedVibe, the recorded speaker and owner of his datasets. Giouli recorded under a signed BedVibe performer agreement. Documentation of performer authorisation is available for private due-diligence review on request. Read the blank performer agreement (V4 template).
Pan — Greek Emotional Speech Dataset
8,160 clips · 5.53 hours · 6 emotions · 48 kHz, 32-bit float. Contact for licensing.
Giouli — Greek Emotional Speech Dataset
8,154 clips · 4.75 hours · 6 emotions · 48 kHz, 32-bit float. Contact for licensing.
One Voice, Eleven Languages
Pan recorded eleven languages, each in all six emotional states, so the speaker stays the same while language and emotion change. English and Greek are complete 1,360-phrase sets of 8,160 clips each. French, German, Japanese, Mandarin, Norwegian, Portuguese, Russian, Spanish and Swedish share a shorter phrase set, recorded in every emotion. Choose a language, then an emotion.
Pan — Polyglot Emotional Speech Dataset
11 languages · 16,861 clips · 11.22 hours · 6 emotions in every language · 48 kHz, 32-bit float. Contact for licensing.
Additional English Voices
Further English voices recorded on the same 1,360-phrase script. Emotional coverage differs from voice to voice: each card lists the states recorded and plays one real sample from each. Licensing on request.
Irini — English Speech Dataset
6,134 clips · 2.17 hours · 6 emotions · 48 kHz, 32-bit float. Contact for licensing.
Jasmine — English Speech Dataset
4,032 clips · 1.19 hours · Neutral, Happy, Whisper · 48 kHz, 32-bit float. Contact for licensing.
Maria — English Speech Dataset
1,937 clips · 0.72 hours · Neutral, Angry, Happy · 48 kHz, 32-bit float. Contact for licensing.
Stelina — English Speech Dataset
1,440 clips · 0.72 hours · Neutral, Happy · 48 kHz, 32-bit float. Contact for licensing.
Parastu — English Speech Dataset
1,409 clips · 0.67 hours · Neutral, Happy · 48 kHz, 32-bit float. Contact for licensing.
Sofia — English Speech Dataset
1,345 clips · 0.56 hours · Neutral · 48 kHz, 32-bit float. Contact for licensing.
Gabriela — English Speech Dataset
1,176 clips · 0.46 hours · Neutral · 48 kHz, 32-bit float. Contact for licensing.
Nina — English Speech Dataset
489 clips · 0.22 hours · 6 emotions · 48 kHz, 32-bit float. Contact for licensing.
License and pricing — what each dataset license lets you do
Four ways to buy: €149 per speaker, €599 for all seven, €1,799 commercial, €4,999 enterprise with sublicensing. Every price is a one-time cost for a perpetual license — no subscription, no per-use fee, no royalty. If you only need to train a model and sell what you generate with it, the €149 or €599 purchase is enough. If you need to distribute the model, embed it in software you ship, or run hosted inference for your own customers, buy the commercial license. If your own customers need rights of their own, buy the enterprise license.
Single Speaker
€149one-time price, per speaker
- One speaker · ~8,158 clips · 6 emotions
- Train models on the audio
- Use the models you train in your own products and client work
- Sell the audio those models generate
- No distributing or selling the trained model itself
- No redistribution of the audio itself
- No SDK or software redistribution
- No hosted inference for third parties
- No sublicensing · no indemnity
Full English Pack
€599one-time price, all 7 speakers
- All 7 speakers · 57,110 clips · 30.69 hours
- Identical 1,360-phrase script across every speaker
- Train models on the audio
- Use the models you train in your own products and client work
- Sell the audio those models generate
- No distributing or selling the trained model itself
- No redistribution of the audio itself
- No SDK or software redistribution
- No hosted inference for third parties
- No sublicensing · no indemnity
Commercial License
€1,799one-time price · includes the full pack
- Everything in the Full English Pack
- Distribute the models you train, commercially
- Embed in SDK, CLI or MCP software you ship
- Run hosted inference for your own customers
- No BedVibe-managed workflow required
- No sublicensing the model or your rights to third parties · Enterprise tier required
- No redistribution of the audio itself
Sublicensable / OEM
€4,999one-time price · includes the full pack
- Everything in the Commercial License
- Sublicense to your own customers
- Performer-rights indemnity for you as licensee — not for your sublicensees — capped at the fee paid
- Audit and field-of-use terms negotiated
- Speaker-disjoint expansion available on request
- Custom recording to your specification
Rights position — why this audio is safe to train on
- Every recording is covered by a signed performer agreement granting BedVibe Studios worldwide, perpetual, irrevocable, transferable and sublicensable commercial rights — including training, fine-tuning, synthesis, licensing and dataset sales.
- Originals are digitally signed through gov.gr, the Hellenic Republic's official digital signature service.
- Contracting entity is BED VIBE GKILIS, Norway, org. no. 935 267 897. Agreements are governed by Norwegian law, with GDPR consent recorded explicitly.
- Derived weights survive. Performers have acknowledged in writing that recordings already used for AI training cannot be withdrawn from trained models, derived features or published outputs. Your models do not become unusable later.
- Redacted clearance attestations — grant clauses visible, personal identifiers withheld — are available under NDA. Originals can be produced for formal counsel review.
- The performer agreement itself is public: read the blank V4 template.
Performer names and personal information
Dataset licenses cover the AI training, fine-tuning, synthesis and commercial-use rights stated for each license tier.
Performer personal information is not part of the dataset license. Where a performer is presented publicly by a first name, stage name or other agreed public name, that name is what the dataset supplies. The license does not grant access to, or rights over, undisclosed legal names, surnames, contact information, addresses, private biographical information, photographs or endorsements.
The source recordings themselves may not be redistributed except where expressly permitted by a separate written agreement.
Verify before you buy — no purchase required
Ask and we send, the same day: the full license text for any tier, and a sample of the
per-clip manifest — audio_file, text, speaker, emotion, language,
gender — with duration distributions per emotion, so you can check same-text
neutral/emotional alignment, performer uniqueness and clip counts before spending anything.
Each of the seven published English sets includes 59–60 non-verbal vocalizations — breathing, panting, gasps, whimpers, screams, crying, laughter, yawns, sneezes, and pain and effort vocalizations.
Beyond the published English sets we hold roughly thirty voices in total. Recorded hours and coverage of emotions and vocalizations vary from voice to voice, and we supply the exact hours, emotional states and material available for any voice on request. Some of the newer recordings were made on a Neumann U87, and new sessions can be recorded to your specification.
Need custom datasets, more hours, new languages, additional emotional states, or fresh recording sessions with professional studio equipment?
Contact bedvibe@bedvibe.studio for custom dataset production and enterprise recording work.