CereProc is known for its emotional speech synthesis and character voice cloning, providing expressive AI voices for entertainment, accessibility and interactive applications. Has been acquired by Capacity and continues to operate as its neural TTS engine.

CereProc

Core parameters and statistics of CereProc

CereProc is a speech synthesis company with a history of more than 20 years. Founded in Edinburgh, Scotland in 2005, it has accumulated a deep technical moat in the field of emotional expression TTS and voice cloning. Acquired by intelligent automation platform Capacity around 2025, its speech engine continues to operate as the core technology of Capacity's neural TTS product line.

Projects	Public Information
Official positioning	Expressive AI voice synthesis / Neural Text-to-Speech
Core technology route	Unit selection synthesis + parametric statistical synthesis + neural TTS
Number of available voices	81 universal voices covering 24 languages and multiple regional accents
Emotional expression	Supports multi-dimensional emotional parameter control such as happiness, sadness, anger, fear, etc.
Sound cloning	Customized sound model training based on a small amount of recording data
Deployment method	Desktop, client/server (cServer), mobile SDK, cloud deployment, local deployment
Place of Attribution	GB (Edinburgh, UK)
Latest version	8.0 (~2025)
Supported platforms	Windows, macOS, Linux, Android, iOS, Web API
SDK Language Bindings	C/C++, C#, Java, Python

Differentiated positioning of emotional speech: CereProc's speech model is not simply "reading words", but introduces emotional labels at the acoustic model layer, so that the same speech can naturally switch between emotional states such as happiness, sadness, anger, fear, etc., and even mix different emotional intonations in the same text. This is essentially different from the "multi-style selection" of mainstream TTS platforms (such as Amazon Polly, Google Cloud TTS) - the latter usually trains models independently for each emotional state and cannot complete emotional transitions within the same sentence.

Historical Status of Sound Cloning: CereProc is an early commercial practitioner of sound cloning technology. The case of reconstructing the voice of film critic Roger Ebert in 2009 and cloning the voice of NFL player Steve Gleason in 2014 is landmark in the history of speech synthesis. This is different from the "instant cloning" routes such as ElevenLabs that have emerged in recent years - CereProc's method focuses more on data quality and fidelity of sound details rather than extremely fast cloning.

Business structure after acquisition by Capacity: CereProc will be integrated into the Capacity platform around 2025, and its TTS engine will be targeted at enterprise customers under the brand name Capacity Neural TTS, while retaining CereProc's own product line. This means that the APIs and services of original customers will not be shut down immediately, but the procurement entrance for new customers will gradually be migrated to Capacity. Existing users relying on this should pay attention to the service continuity terms.

User and market recognition of CereProc

CereProc's market recognition presents a two-tier structure of "deep reputation in the barrier-free field + extensive coverage of enterprise call centers through Capacity".

Iconic Status in Accessibility: CereProc, a practice that provides voice cloning services to ALS/MND patients, has built strong brand trust within the medical and accessibility communities. Roger Ebert used CereProc to synthesize a voice speech ("Remaking My Voice") at TED2011, and Steve Gleason used a cloned voice to appear in Microsoft's 2014 Super Bowl commercial. They are two milestones in TTS technology's empowerment of humanistic care. The 2011 documentary "Giving the Critic Back His Voice" produced by BBC Radio 4 fully documented the entire process of CereProc's reconstruction of Ebert's voice.

Enterprise-level customer coverage: Through the Capacity platform, CereProc’s neural TTS engine serves 20,000+ organizations, covering well-known global brands such as 3M, Disney, LinkedIn, Nike, Sony, NVIDIA, American Express, Pearson and more. These customers primarily use TTS for IVR self-service, call center, and automated outbound call scenarios.

Industry Evaluation and Benchmarking: CereProc has received positive reviews from third-party evaluations in terms of naturalness of emotional expression and accuracy of regional accents. However, in recent years, with the rise of new generation TTS platforms such as ElevenLabs and PlayHT, public benchmark ranking data no longer has the advantage. The specific MOS (average opinion score) comparison data is based on the official page and third-party reviews.

Limitations: Compared with emerging platforms such as ElevenLabs, CereProc is less well-known on the C side and lacks an online trial platform and lightweight subscription plan for individual users. The API call volume and ecological integration richness in the developer community are also inferior to Amazon Polly and Google Cloud TTS.

Cost Advantages of CereProc

The cost structure varies depending on the usage method: C-side users can usually experience core functions through the free version, and high-frequency usage requires subscribing to paid packages; developers/API users are billed based on the number of calls; enterprise-level users need to contact the business to obtain customized quotations. The specific price is subject to the official real-time pricing page.

Main functions of CereProc

CereProc 的功能体系围绕"语音自然度 + 情感表达 + 部署灵活性"三条主线展开，不是简单的文本朗读工具，而是一套可深度集成的语音引擎。

• Multidimensional Emotional Speech Synthesis: Control the emotional state of speech through SSML tags or API parameters, supporting basic emotions such as happiness, sadness, anger, fear, surprise, disgust, etc., and can adjust emotional intensity. Key difference: Emotional parameters can be switched at sentence granularity - in the same conversation, after character A finishes speaking in an angry tone, character B immediately responds in a calm tone, without switching the speech model. This has direct application in audiobook dubbing and interactive game character dialogue.

• Voice Cloning and Customized Sound: Train a customized sound model based on the recording data of the target person. The amount of data required depends on the engine type - the recording requirements of the parameterized engine are much less than that of the unit selection engine. Acceptance Concern: The naturalness of a cloned sound depends largely on the quality and diversity of the source recording; if there is background noise, uneven volume, or emotional monotony in the recording, the cloned result will amplify these flaws. It is recommended that 30-60 minutes of diverse recording samples be provided for evaluation prior to delivery.

• Dual Engine Architecture: Provides both unit selection (Unit Selection) and parametric (Parametric/Neural) synthesis engines. The unit selection engine splices the best segments from the pre-recorded speech library, which is more natural but takes up a lot of space; the parametric/neural engine generates waveforms based on the acoustic model, which takes up a small space and supports dynamic emotional control. Implementation Tips: For standard TTS scenarios, it is recommended to give priority to using the neural engine to balance naturalness and deployment cost; in scenarios that require extremely high pronunciation accuracy (such as medical terminology reading), the unit selection engine may be more reliable due to "human mouth type".

• Multi-platform SDK and API: Provides SDK bindings in four languages: C/C++, C#, Java, and Python, covering the five major platforms of Windows, macOS, Linux, Android, and iOS. Integrates with third-party systems via gRPC and MRCP protocols, especially suitable for replacing legacy TTS engines in legacy IVR systems.

• Regional accent refinement: Taking English as an example, CereProc provides accent options such as American English, Southern England English, Northern England English, Scottish English, Glasgow English, Lancashire English, Irish English, Welsh English, and West Midland English. It also provides regional difference coverage such as French (Paris/Quebec), German (Standard/Austria), and Norwegian (Written/New Norwegian). This kind of accent accuracy is rare in large-scale TTS platforms.

• Special effects voices: Provides special effects voices such as Demon, Ghost, Goblin, Pixie, Robot, etc., which can be used for game character dubbing and entertainment content creation.

CereProc model and version evolution

CereProc's product iterations are based on the engine version number (CereVoice Engine), and important technological leaps occur every 2-3 years.

Unit selection foundation period (2005-2018)

• The establishment of CereProc and the first-generation engine (2005): The company was founded in Edinburgh. The first-generation CereVoice engine was based on unit selection synthesis technology and synthesized speech through splicing of large-scale recording databases.
• CereProc Sound Cloning Milestone (2009-2014): Provided voice cloning services to Roger Ebert (2009) and Steve Gleason (2014), validating the practical value of custom sounds in accessibility scenarios.
• cServer Enterprise Edition launched (about 2015): The client/server product for the IVR market is launched, supports Windows and Linux deployment, and becomes the main revenue driver of enterprises.

Deep learning transformation period (2019-2023)

• CereProc 5 (approximately 2019): For the first time, deep learning speech synthesis capabilities are introduced, and neural network components are introduced into the acoustic model to improve the naturalness of speech.
• CereProc 6 (about 2021): The acoustic model architecture is upgraded, the language coverage is expanded to 24, and mobile SDK support is added.
• CereProc 7 (circa 2023): Enhanced Neural TTS engine becomes the main feature, emotion control function is moved from experimental to standard capability. The sound cloning efficiency of the parametric engine is significantly improved - the amount of recording data required is reduced by approximately 50% compared to the previous generation.

Acquisition and integration period (2025-present)

• CereProc 8 (~2025): Latest version after acquisition by Capacity. The naturalness of emotional speech and the quality of voice cloning are further improved, and the engine is simultaneously launched to the market under the Capacity Neural TTS brand. New customers' independent CereProc product purchases are gradually migrating to the Capacity platform.

Version History Summary: CereProc started from traditional unit selection and synthesis, and entered the neural TTS era after deep learning transformation. The acquisition in 2025 marks a new stage of its transition from independent operation to "technical component + platform integration". The precise date of version number iteration has not been made public. The above time points are based on publicly available product milestones.

CereProc’s technical advantages

CereProc's technology stack has established quantifiable differentiation in emotion control and regional accent fidelity, but it also faces pressure to upgrade its technical route under the "single-mouth" trend of end-to-end neural TTS.

Engineering significance of dual-engine hybrid strategy: The coexistence of unit selection engine and parameterization engine is not a backward technology, but a pragmatic choice for different scenarios. The unit selection engine directly splices speech segments from the real-person recording library, which is naturally superior to purely generative models in terms of pronunciation accuracy and accent authenticity - because each segment is pronounced by a real person, there is no "pronunciation blurring" problem common in neural TTS. The price is that the voice library can often be hundreds of MB or even GB in size, and the emotion cannot be adjusted in real time. The parametric/neural engine achieves flexible emotional control with a smaller model size (MB level), but its performance on rare vocabulary and complex accents is not as stable as unit selection. The coexistence of two engines means that CereProc's SDK needs to maintain two sets of synthesis pipelines at the same time, which places certain requirements on the development and testing work of the integrator.

Mechanism of emotion-aware acoustic modeling: CereProc introduces emotion labels as conditional inputs in the acoustic model training phase, allowing the model to learn to produce corresponding timbre, intonation and rhythm changes in different emotional states. Different from the "post-processing tuning" route commonly used by competing products, CereProc's emotional control occurs in the acoustic feature generation layer, so there will be no timbre mutation or mechanical feeling when switching emotions. The technical cost of this approach is that the training data needs to be labeled with emotional categories sentence by sentence, and the data preparation cost is high.

Data Efficiency Route for Sound Cloning: CereProc's Sound Cloning optimizes data requirements - the parametric engine requires only 15-30 minutes of recording data to produce a recognizable custom sound, far less than the hours of recording required by the unit selection engine. However, for customers pursuing high fidelity (such as sound reproduction in the media industry), it is still recommended to provide 2-3 hours of multi-scene recording data to cover sufficiently rich pronunciation variations.

Cross-platform consistency of SDK architecture: CereProc's SDK adopts a unified C language core + each platform binding layer architecture to ensure consistent synthesis effects on Windows, macOS, Linux, Android, and iOS. The design of the gRPC interface allows developers to independently deploy the TTS engine as a containerized service in a microservice architecture and call it through standard protocols without binding a specific programming language.

How to use CereProc

The usage paths of CereProc vary significantly depending on the target scenarios - individual users and enterprise customers have completely different entry points.

Desktop Voice Installation (Individual Users): Purchase the desired voice (by accent/language) at cereproc.com and download the installer to Windows or macOS systems. After the installation is complete, the voice will be automatically registered as the system voice and can be used in any application that supports the system TTS API (such as screen readers, document reading tools). Limitations: Price and availability for non-English voices must be confirmed separately, and online trials are not supported.

Sound Clone Service (Custom Sound): Submit your custom sound request through the CereProc website or Capacity sales team. The process usually is: provide recording samples → assess data quality → sign a training contract → model training (takes several weeks) → deliver voice packages. Training period and cost depend on the type of engine required (unit selection is more expensive and slower, parameterization is faster and cheaper) and the speech complexity of the target language.

Enterprise API/cServer Integration: Enterprise customers request a demo or contact sales through Capacity’s TTS page (capacity.com/text-to-speech-software/). Integration methods include:

• gRPC API: suitable for modern microservice architecture, supports streaming composition, and has low latency
• MRCP protocol: Compatible with traditional IVR and contact center platforms, and can replace old TTS engines in Genesys, Avaya, Cisco, etc.
• Local deployment (cServer): Run the TTS engine on your own server, suitable for sensitive data sovereignty or offline scenarios

Quickly integrate concept code (Python):

# CereProc SDK example (pseudocode, the actual interface is subject to the official SDK document)
import cerevoice

engine = cerevoice.Engine(
    license_key="<YOUR_LICENSE_KEY>",
    voice="cerevoice_heather_22k" # Scottish English voice
)

# Emotional Control: Read aloud in a happy tone
output = engine.speak(
    text="I'm so excited to tell you about this!",
    emotion="happy",
    emotion_intensity=0.8
)

Notes before integration: CereProc's SDK requires a valid License Key to run. During the evaluation phase, you can apply for a trial license from sales. The licenses of different voice packages are independent of each other. If the project requires multiple accents/languages, you need to confirm the Bundle authorization plan in advance. Streaming synthesis of the gRPC interface requires the client to process the splicing and buffering of audio streams in long text scenarios.

Product Pricing for CereProc

CereProc's pricing system is based on project-based and enterprise contracts, and lacks a public self-service subscription page. This is the biggest difference in commercialization between it and emerging TTS platforms.

Desktop Voice Purchase: The desktop license price for a single voice is not disclosed. Referring to TTS industry practice, the price of a single voice is usually in the range of $30-$100, but CereProc's English voice may have a premium in pricing due to differences in accents (for example, Scottish English and Welsh English are "niche accents").

Sound cloning customization: Customized sound training adopts a project-based quotation, and the cost is affected by the following factors: the speech complexity of the target language (such as Chinese Mandarin vs English), engine type (unit selection > parameterization), recording data quality (whether a professional recording studio is required), and delivery cycle. Based on industry experience, the cost of customizing a parametric engine is in the range of $5,000-$15,000, and a unit selection engine may be as high as $20,000-$50,000. Risk Warning: The voice package generated after training is usually bound to a specific engine version, and additional fees may be required for engine upgrades.

Enterprise cServer License: Billed based on the three-factor combination of the number of concurrent channels + the number of audio channels + the number of voices. The contract period is usually one or three years. Since it is sold through Capacity, enterprises can also choose Capacity's SaaS model (billed by interaction volume), which means that the actual cost of using CereProc TTS may be packaged through the Capacity platform's turnkey contract and is difficult to separate separately.

Purchase Suggestion: Before obtaining a formal quotation, it is recommended to clarify the following terms with the salesperson - the privacy protection and deletion policy of training data, the limitations of the commercial use scope of the voice package (such as whether it can be used for TV advertising/streaming media), the attribution of migration costs for engine version upgrades, and the continuation of the right to use the voice package after the contract is terminated. These terms vary significantly between providers, and policies may be changing following CereProc's acquisition by Capacity.

Application scenarios of CereProc

CereProc's implementation scenarios focus on vertical fields that require "personalized voice" and "emotional expression" rather than general text reading.

• Accessible Assistive Communication: This is CereProc’s most socially valuable scenario. Preserve "their own voice" for patients with speech disorders caused by ALS/MND, laryngeal cancer, etc., and output personalized voices on assistive communication devices through voice cloning technology. Actual benefits: Patients can record a small number of voice samples before their condition worsens, generate a digital voice copy and use it for daily communication, maintaining the continuity of their personal voice and sense of identity. Implementation method: Achieved through CereVoice Me service or customized solutions from hospitals/rehabilitation institutions.

• Entertainment and Media Dubbing: Game character dubbing requires emotional variation and accent differences, and CereProc's 81 voice libraries and special effects voices can be used for game NPCs, animation dubbing, and audiobook production. Quantitative comparison with actual human voice dubbing: Using CereProc synthetic dubbing can compress the recording time of a single character from hours to minutes, but the emotional expressiveness still cannot completely replace the performance of professional voice actors under extreme emotions (such as breaking down and crying, laughing wildly). Suitable Scenarios: Small and medium-sized game projects with limited budget, audio content that needs to be quickly produced in multiple languages.

• Enterprise IVR and Customer Service Center: Through the Capacity platform, CereProc's neural TTS engine provides natural voice announcements for call centers. Dramatically reduce customer hang-ups during self-service processes compared to legacy TTS. Implementation Tips: IVR scenarios have higher requirements for voice stability than emotional expressiveness. It is recommended to prioritize the use of neural engines and complete A/B testing (old TTS vs CereProc new voice) before going online to quantify changes in hangup rates.

• Brand Sound Asset Creation: Enterprises can customize exclusive brand sounds for use in advertisements, product introduction videos and social media content to form a unified sound recognition system. Actual Case: After a global brand used CereProc to customize its voice, the brand hearing consistency of its telephone customer service was optimized, but the specific data is subject to the official case.

• Education & Language Learning: CereProc's 24-language coverage and multiple regional accents can be used for standard pronunciation demonstrations in language learning applications. Boundary Note: CereProc's accent accuracy (for languages ​​with fewer resources, such as Scottish Gaelic and Welsh) is a unique value, but in mainstream languages ​​(American English, Mandarin), its pronunciation naturalness is on the same level as Google Cloud TTS and Microsoft Azure TTS, with no obvious advantage.

Applicable groups of CereProc

Due to its historical and technical positioning, CereProc’s user pedigree is concentrated on the B-end and groups with special needs, and the threshold for reaching individual C-end users is relatively high.

• Accessibility needs and medical/rehabilitation facilities: ALS/MND patients, post-surgery laryngeal cancer patients, stroke survivors with speech impairments, and non-profit organizations and hospitals that provide support to these populations. Core Value: Preserving the user's own voice instead of using a universal voice, which is far superior to standard TTS in terms of psychological identification and communication effects. Unsuitable Boundary: Voice cloning requires patients to have clear recording conditions. For patients who have lost their ability to pronounce sounds, they cannot be reconstructed reversely, and their voices can only be "donated" by family members or close people.

• Game Development and Media Production Team: Small and medium-sized studios that need to dub characters but do not have the budget to hire professional voice actors, or that need to quickly produce multi-language dubbing projects. Implementation Tips: It is recommended to use CereProc for daily conversations and system broadcasts of non-player characters (NPCs). The protagonist and key plot dialogues still use real-person dubbing to ensure emotional tension. Projects that require special effects voices (demons, robots, etc.) can benefit directly from CereProc's FX voice library.

• Enterprise IT and Customer Experience Team: The enterprise team responsible for call center technology selection, IVR system upgrades or brand voice projects. Purchase Prerequisite: The enterprise is already using or planning to deploy the Capacity platform, or needs an on-premises TTS engine to meet data compliance requirements. Not suitable for the boundary: If the enterprise only needs basic cloud TTS and does not have strong needs for emotional expression, Amazon Polly or Google Cloud TTS is better in terms of cost and integration convenience.

• Voice Application Developer: Developers who integrate CereProc TTS into their own applications through the SDK. Suitable for application scenarios that require offline TTS capabilities, emotional parameter control, or regional accents. Technical threshold: Developers need to understand the differences between the dual engines and make logical branches at the code level. During the evaluation period, you need to apply for a trial license and involve business processes. It is not suitable for individual developers to do prototype verification.

Summary and Outlook

CereProc has more than 20 years of technology accumulation in the field of emotional speech and voice cloning. Its product combination of "accent refinement + controllable emotional parameters + dual-engine architecture" has formed a unique ecological niche in the TTS industry. After being acquired by Capacity, the TTS engine has gained a larger enterprise customer channel and more complete intelligent automation platform support, but it also means that the pace of marketization of independent products may slow down.

Current core advantages: The breadth of regional accent coverage (especially the dialects across the British Isles) is irreplaceable in mainstream TTS platforms; the depth of the emotion control mechanism (acoustic model layer rather than post-processing layer) is better than most competing products; brand recognition and high trust in sound cloning in the barrier-free field; local deployment solutions meet strict data sovereignty requirements.

Current main limitations: C-side availability is extremely low - no free trial layer, no online demo, no self-service subscription entrance, almost inaccessible to individual users; API prices are opaque and quoted on a project basis, making selection costs high for small and medium-sized developers; the product route after acquisition is unclear, and the long-term availability of independent APIs is uncertain; although the number of supported languages (24) is more than that of small and medium-sized TTS vendors, it is far less than Amazon Polly (80+ languages) and Google Cloud Coverage of TTS (100+ languages); end-to-end neural TTS With the "one-stop-clear" trend, the maintenance cost of dual engines may gradually be higher than that of a single end-to-end engine.

Follow-up observation points: The depth of Capacity's technology integration with CereProc - whether to maintain dual-brand operations or gradually merge into the Capacity unified platform; whether CereProc's independent API will continue to accept new customer registrations; whether the delivery cycle and pricing of the sound cloning service will change under the Capacity system.

Procurement and Adoption Risk Assessment: For accessibility scenarios, CereProc’s voice cloning quality and usage history are credible, but it is recommended that the deletion terms of training data and the permanent use rights of voice packages be made clear in the procurement contract. For media and game dubbing scenarios, it is recommended to apply for a short-term trial on the Capacity platform first, and use actual project data to evaluate the synthesis effects and delivery costs before signing an annual contract. For projects that require standard TTS and have no strong requirements for emotional control or specific accents, give priority to comparing the API solutions of Amazon Polly and ElevenLabs - the former is lower cost and the latter has a better C-side experience. Any long-term procurement involving CereProc should include a "service continuation and data migration after product route change" clause in the contract to hedge the risk of possible product strategy adjustments after the acquisition.

Related tools: elevenlabs, udio

How to use CereProc

• Web client: You can use it by visiting the official website and registering an account. Most functions do not require installation.
• API Access: Provides RESTful API, developers can obtain the API Key and integrate it into their own applications.

Version Info

• CereProc 8 ：There is no official precise date yet; the latest version of the engine improves the naturalness of emotional speech and the quality of sound cloning. （~2025）
• CereProc 7 ：No official precise date yet; introducing enhanced neural TTS and emotion control functions. （~2023）
• CereProc 6 ：There is no official precise date yet; new acoustic model upgrades and more language support have been added. （~2021）
• CereProc 5 ：There is no official precise date yet; deep learning speech synthesis capability will be introduced. （~2019）