AIxBlock

What is AIxBlock?

AIxBlock is a partner that provides training data for enterprise voice AI and large language model teams, offering services such as data collection, transcription, annotation, quality control, model evaluation, and self-hosted delivery.

The current official website focuses on voice, audio, text, and dialogue data, and no longer positions itself as a general decentralized AI development platform as mentioned in earlier promotions. When making purchases or evaluations, it is necessary to rely primarily on the existing product pages and project contracts.

Main services of AIxBlock

1. Collection of voice and dialogue data

AIxBlock can organize professional voice actors or ordinary native speakers to record script audio, natural conversations, and audio content for specific business scenarios, covering various languages, accents, ages, and demographic groups.

• Script reading and spontaneous conversation recording.
• Switching between accents, dialects, and language codes in different regions.
• Call centers, voice assistants, and customer service scenarios.
• Custom data for ASR, STT, TTS, and conversational AI.

2. Transcription and speech annotation

The team can obtain word-for-word transcriptions, timestamps, separation of speakers, phoneme labeling, as well as labels indicating emotion and intent. The project defines annotation guidelines and acceptance criteria based on the objectives of the model.

Voice data often contains overlapping speech, background noise, colloquial expressions, and incomplete sentences; therefore, it is necessary to decide in advance whether to retain these authentic features before starting the project.

3. Environmental and industrial sound data

In addition to human voices, AIxBlock can also capture sounds from streets, offices, natural environments, machines, devices, household appliances, alarms, and other sound events, for use in acoustic classification and anomaly detection.

• Sound event detection and environmental scene recognition.
• Identification of abnormal sounds in industrial equipment.
• Training of noise reduction and echo processing models.
• Development of security systems, smart home solutions, and assistive devices.

4. Text and dialogue data

Services tailored for large language models include multi-turn dialogue annotation, intent classification, entity recognition, slot filling, and domain-specific text collection. The data can be used for training, fine-tuning, evaluation, and security testing.

5. SFT and RLHF preference data

AIxBlock can generate supervised fine-tuning data by pairing prompts with corresponding responses, and it can also enable people to compare the outputs of various models, sorting them based on accuracy, security, and task completion.

Such data should be designed in line with specific business rules, rather than merely collecting general positive or negative feedback. In fields such as healthcare, law, and finance, the involvement and review of relevant experts are also necessary.

6. Ready-made call center audio library

The authorities provide authentic call center audio files that can be licensed directly, covering various languages, accents, and business sectors. Companies can first obtain samples to assess the acoustic characteristics and data format before deciding to acquire licenses in bulk.

Some data can be provided along with transcription and annotation. When it comes to names, payment cards, or other personal information, the platform will anonymize such PII as needed, but the purchaser still needs to verify the specific criteria for handling this data.

7. Self-hosted data platform

AIxBlock enables the deployment of workflows for data collection, annotation, quality control, training, and evaluation in the customer’s on-premises environment, private cloud, or hybrid setup. Customers can connect their own storage systems from the beginning of the project.

This approach helps to keep the original data within the customer’s secure environment, reducing the risk of third-party SaaS services storing proprietary data for long periods; it is suitable for banks, healthcare institutions, governments, and other regulated organizations.

Quality control methods

• Multi-layer verification of identities, devices, and sessions helps to reduce cases where proxy tasks do not correspond to the correct identities.
• Consistency metrics, consensus checks, and random sampling by senior reviewers are used for annotation.
• Use the project-specific AI assistant to answer questions regarding labeling rules.
• Use automated quality monitoring to identify abnormal and inconsistent results.
• Delivery, sampling, and rework are managed by the project manager and domain experts.

A quality mechanism can only increase the likelihood of detecting problems; it cannot replace the customer’s own benchmark tests. Companies should specify in the contract the accuracy rate, sampling proportion, types of errors, rules for rework, and methods for final acceptance.

Which users is it suitable for?

• A team responsible for developing speech recognition, speech synthesis, and voice assistants.
• Companies that need training data in multiple languages or less common languages.
• Companies that train dialogue models for customer service, banking, and call centers.
• Large-model teams that carry out SFT, RLHF, DPO, or model evaluation.
• Sensitive training data cannot be uploaded to public SaaS platforms in regulated organizations.
• We are seeking contributors who wish to participate in tasks such as recording, transcription, language annotation, or AI evaluation.

Corporate procurement process

1. Define the model objectives, data types, language, region, audio requirements, and the final delivery format.
2. It explains the project scale, timeline, domain expertise, privacy level, and boundaries for data usage.
3. Apply to AIxBlock for samples, catalogs, or quotes for custom projects, and sign the necessary confidentiality agreements.
4. Verify the data source, participant consent, scope of authorization, exclusivity, and commercial use rights.
5. Establish labeling guidelines, gold standards, quality metrics, as well as rules for sampling and rework.
6. Decide whether to use a managed delivery service or a self-hosted platform, and conduct security and architecture reviews.
7. First, run a small-scale pilot to verify the performance and any deviations on your own model.
8. After meeting the criteria in the pilot phase, expansion is carried out in stages, with continuous monitoring of data drift, language coverage, and model performance.

Service versions and pricing methods

As of August 25, 2026, AIxBlock has not released a unified pricing schedule based on rows, hours, or seats. The official website uses an inquiry-based system for pricing; the costs are determined depending on the type of data, language, scale, expert level, quality requirements, delivery speed, and method of deployment.

The project size options listed on the official website’s contact form range from less than $10,000 to over $5 million; these are merely classifications of requirements, not the prices of the packages or the minimum order amounts. The final price shall be determined based on the written quote, the scope of work, and the contract.

Why isn’t there a fixed price for each item?

General intent labeling and medical conversations reviewed by clinical experts involve completely different costs, risks, and quality requirements. Voice-related projects are also affected by the recording environment, the rarity of certain accents, the level of detail in transcription, and privacy considerations.

Purchasers should not consider only the price of individual items; they must also take into account the overall costs associated with quality rates, rework, project management, certification, as well as storage and security audits.

Data authorization and copyright verification

The value of training data depends not only on its quantity but also on whether there are clear consent forms for collection, permits for use, and guidelines regarding retraining. In the case of recordings from actual call centers, it is particularly important to verify the original statements provided, as well as the rules governing the handling of personal information and its cross-border transmission.

• Determine whether the data is subject to a custom exclusive license, a non-exclusive license, or a re-licensable license.
• Clarify whether it can be used for training, fine-tuning, evaluation, commercial deployment, and redistribution.
• It is required to keep the participants’ consent and proof of data source.
• Specify the procedures for deletion, withdrawal, correction, and handling of rights complaints.
• Avoid interpreting “desensitized” as meaning it cannot be identified again at all.

Privacy and security

The AIxBlock privacy policy covers personal data of website users, customers, project contributors, and those involved in the services it offers; it also references frameworks such as GDPR, the UK Data Protection Act, and CCPA. Cross-border data processing may rely on standard contract clauses or other legal mechanisms.

The official security page states that the platform has obtained ISO 27001:2013 certification and SOC 2 Type II compliance. Companies should still request the currently valid certificates, the scope of the audits, any exceptions, and a list of subcontractors, rather than relying solely on the website logo.

• Define the roles of data controllers, processors, and subcontractors.
• Review identity permissions, logs, keys, backups, and incident response.
• Separate controls are applied to PII, PHI, payment information, and government data.
• Check which components in a self-hosted environment still need to be connected to external services.
• After the project is completed, verify that the data copies, caches, and temporary files have been deleted.

Ways for contributors to get involved

AIxBlock is also seeking audio recordists, transcribers, language experts, domain specialists, and AI evaluators. Contributors can apply for projects, complete tests, and carry out the work required by those projects.

Before participating, it is necessary to check the details of the work, the purpose for which the data will be used, the payment method, tax implications, identity verification procedures, and the conditions under which participation can be withdrawn. When dealing with biometric data, health information, racial data, or other sensitive information, the platform must obtain explicit consent in advance.

Open-source status and GitHub details

Around 2025, AIxBlock announced that it would make its general-purpose AI development platform available under an open-source license, and it made the relevant GitHub repositories public. Upon verification, both of the official repositories mentioned in those announcements are no longer accessible; they return a page indicating that the repository does not exist.

Therefore, the current enterprise training data services and self-hosted products cannot be simply labeled as \"active open-source projects available for download from GitHub.\" Historical open-source code, current commercial platforms, data assets, and customer deliverables need to be assessed separately.

Product advantages

• It focuses on voice, audio, text, and dialogue data; it is not a general-purpose annotation platform.
• It supports over 100 languages and various regional accents.
• It covers the entire process, including data collection, transcription, annotation, preference data handling, and quality control.
• Ready-made audio libraries can shorten the time from procurement to model experimentation.
• Self-hosted delivery is suitable for organizations with requirements regarding data sovereignty and auditing.

Usage restrictions

• There is no fixed public price; initial steps involve business discussions, sample testing, and contract negotiations.
• The accuracy rates and delivery times shown on the official website are based on specific projects, and it is not guaranteed that they will be the same for other projects.
• Self-hosting can reduce the risk of data leakage, but it increases the customer’s responsibilities regarding their own infrastructure and operations.
• Data quality and model performance are still influenced by annotation standards, language coverage, and training methods.
• Historical product information may not be consistent with the current orientation of the company’s data services.
• The old GitHub repository is currently inaccessible, so it is not possible to determine whether there is still an open-source version that is under maintenance.

Frequently Asked Questions

What is AIxBlock mainly used for now?

Currently, it offers enterprise-level voice, audio, text, and dialogue training data, as well as ready-made call center audio and self-hosted data platforms.

Is it possible to purchase only ready-made data?

It is possible to request ready-made audio catalogs and samples; transcription and annotation services can also be added as an option. The scope of authorization, languages, formats, and prices need to be confirmed separately.

Is RLHF data supported?

Supported. The platform can bring together team members to compare the model responses, and rank and score them based on the accuracy of the project definition, its security, and the degree of task completion.

Does self-hosting mean having no external dependencies at all?

Not necessarily. The purchaser still needs to verify whether authentication, user access, models, the GPU market, as well as update and support components are connected to external services.

Is AIxBlock an open-source tool?

In the past, announcements regarding open-source status were made, but the corresponding official GitHub repositories could no longer be accessed upon verification. At present, enterprise services should be regarded as commercial projects; any claim of open-source status requires the existence of new official repositories along with appropriate licensing documentation.