G-SODA is the global product brand of Timehill Inc., Japan.

Rights Cleared for Commercial AI

All participants sign a legally binding agreement confirming that:

* their recordings contain no personally identifiable information (PII);
* all speaker metadata is accurate;
* their speech recordings may be used, licensed, sublicensed, distributed, and commercially exploited worldwide as part of the G-SODA Corpus (THCP), in perpetuity, for AI training, machine learning, research, and commercial applications.

Time Hill Speech Corpus (THCP)

Product Overview

Time Hill Speech Corpus (THCP) is a collection of professionally recorded, richly annotated Japanese speech resources ready for immediate licensing.
Similar corpora in other languages can be supplied on request.
THCP supports advanced research and commercial applications in automatic speech recognition (ASR), text-to-speech (TTS), speaker diarization, intent understanding and conversational AI.

1 Standard Japanese Set

1.1 Datasets Code

★ Recorded in a transparent booth for each speaker

Code Scenario Participants Duration
(Hours)		 Speakers
1 THCP‑CDSJ25 Face‑to‑face dialogues ★ 2 2,800 hrs 1,500
2 THCP‑CCSJ25 Face‑to‑face group dialogues ★ 3 100 hrs 300
3 THCP‑CKSJ25 Child dialogues (child–child /child–parent) ★ 2 50 hrs 80
4 THCP‑CHSJ25 Medical dialogues (doctor/nurse–patient) ★ 2 50 hrs 30
5 THCP‑CPSJ25 Simulated mobile telephone dialogues ★ 2 150 300
6 THCP‑CTSJ25 Text to Speech (Novel, Basic Sentence, 4-degit number, etc) 1 300 hrs 1,000
7 THCP‑CMSJ25 Monologue Speech(My Life Story, Most Exciting Event in My Life, etc) 1 300 hrs 1,000
8 THCP‑CLSJ25 Lecture Speech (Professor of University) 1 300 hrs 2

1.2 Deliverable Specifications

  • ● Format: WAV 48 kHz / 16-bit, 24bit, 32bit
  • ● Alignment: 1 ms precision start/end times
  • ● Annotation: time-aligned transcription

1.3 Demographics

Regions represented:
Hokkaidō, Tōhoku, Kantō (incl. Tokyo), Chūbu, Kansai (Osaka/Kyoto), Chūgoku, Shikoku, Kyūshū & Okinawa.

Code Gender ≈ F/M Age Range
1 THCP‑CDSJ25 80/20 % 16–90 yrs (90 % 20s–60s)
2 THCP‑CCSJ25 80/20 % 16–90 yrs (90 % 20s–60s)
3 THCP‑CKSJ25 70/30 % 06–15 yrs
4 THCP‑CHSJ25 95/05 % 25–60 yrs
5 THCP‑CPSJ25 50/50 % 16–90 yrs (90 % 20s–60s)
6 THCP‑CTSJ25 70/30 % 16–90 yrs (90 % 20s–60s)
7 THCP‑CMSJ25 70/30 % 16–90 yrs (90 % 20s–60s)
8 THCP‑CLSJ25 00/100 % 60-80 yrs

1.4 Pricing (First Time License / Speech data without annotated transcription)

Transcription and annotation data can be tailored to each customer’s specific requirements, with details to be discussed separately based on the content.
Volume and returning customer discounts are available.

Code List Price (per hrs)
1 THCP‑CDSJ25 $300.00
2 THCP‑CCSJ25 $500.00
3 THCP‑CKSJ25 $400.00
4 THCP‑CHSJ25 $600.00
5 THCP‑CPSJ25 $300.00
6 THCP‑CTSJ25 $150.00
7 THCP‑CMSJ25 $200.00
8 THCP‑CLSJ25 $250.00

1.5 Licensing & Delivery

  • ● Licence type: Non-exclusive, perpetual use within the licensee’s organisation; duplication for internal 　model training, fine-tuning and evaluation is allowed.
    Resale or transfer to third parties is strictly prohibited.
  • ● Delivery lead-time: Approximately one week after contract execution.
    Files are supplied via a secure Dropbox link.
  • ● Custom transcripts/annotation: Quoted separately once the desired specification is agreed.

1.6 Compliance & Child/Medical Data

All child (6–12 yrs) and medical dialogs are fully compliant with APPI, GDPR and HIPAA like legislation: parental/guardian consent is documented; personal identifiers are anonymised or pseudonymised.

2 Why THCP?

2.1 Inventor & Vision

THCP is created by Yoichi Tokioka- novelist, serial entrepreneur and former language research scholar at Waseda University, Tokyo. Since the early 1980s, Mr Tokioka has worked on corpus driven approaches to natural language processing.
His earlier corpora was adopted by Microsoft (Seattle). 

2.2 Rigorous Speaker Recruitment

We consider speaker adoption to be one of the most important aspects.
We believe truly natural speech cannot be captured if speakers memorize or rehearse scenarios. Unlike conventional corpora that accept “any willing native”, THCP follows strict casting rules:

  • ● Authentic roles – Flight attendant dialogs are voiced by real flight attendants; café scenarios by actual baristas, and so on.
  • ● Genuine relationships – Doctor–patient, parent–child and sibling dialogs feature real participants.
  • ● Wide demographics – We include infants (80 yrs), medical professionals and speakers offering first hand accounts of the pre WWII era. No scripts or rehearsals; proprietary facilitation yields genuine overlaps, fillers and hesitations.

2.3 Studio Grade, Multi Track Recording

All sessions are captured in Time Hill’s own acoustically treated studios:

  • ● Independent booths with natural crosstalk monitoring
  • ● Pro grade microphones & interfaces
  • ● 48 kHz / 16,24,32 bit WAV
  • ● Clean channel per speaker—no cross talk contamination

Our setup lets two speakers see each other’s faces while conversing from separate soundproof booths. Each hears the other through headphones, allowing for natural interaction and independent recording of overlapping speech.

Our phone setup places speakers in separate soundproof booths, unable to see each other, simulating real phone calls. They communicate through headphones, and overlapping speech is recorded on separate tracks.

Our setup for group conversations. The setup allows the speakers to be visually connected through transparent partitions enabling natural face-to-face interaction while remaining acoustically isolated. Each speaker is recorded on an individual mono audio track via synchronized, high-quality microphones, ensuring perfectly time-aligned recordings. This configuration allows clear separation of overlapping speech and eliminates background noise, resulting in a clean, precisely synchronized multi-speaker speech corpus.

3 Beyond our Standard Japanese Set

Need something unique? Our team can design and record bespoke corpora in >50 languages with options such as:

  • ● Text to speech prompts: numerals, proper nouns, daily sentences
  • ● Domain specific speech: travel & tourism, device tutorials, sports commentary, emotion rich utterances, accessibility speech (hearing /vision impaired), pathological speech, animal sounds, environmental ambience and more.
  • ● Singing voice recordings: solo, duet or choir tracks for singing.
  • ● Simultaneous-interpretation audio (e.g., Japanese ↔ English): two or four-channel capture of interpreter + speaker in real time

★ Contact

PAGE TOP