ONDEWO Speech-to-Text
ONDEWO Speech-to-Text (S2T) Platform is an enterprise voice-to-text transcription platform from the Austrian vendor ONDEWO GmbH. It runs on-premise via Docker or in ONDEWO's cloud, integrates over gRPC, and trains customer-specific models through transfer learning.
What is ONDEWO Speech-to-Text?
ONDEWO Speech-to-Text (S2T) Platform is described by its publisher as a software platform for enterprises to transcribe human voice in form of audio to text. It is one of the six modules of the ONDEWO suite, alongside CCAI (Call Center AI), NLU, T2S, VTSI and AIM, and it comes from ONDEWO GmbH, a company based in Vienna, Austria and led by its founder Dipl.-Ing. Dipl.-Ing. Dr. techn. Andreas S. Rath.
The selling point ONDEWO puts forward is a combination of highly customer-specific models, very high transcription speed and multilingual coverage, all built on top of general and industry pre-trained models. The published figures are unusually precise for this market: recognition accuracy of 85-95% using recent deep learning algorithms, transcription 40 to 60 times faster than cloud providers, and word error rates of 10-15% on general models falling below 10%, which the vendor calls human level, on customised ones. Ten languages are supported as standard, although only four are actually named: German, English, Spanish and French. Dialects and non-native speakers are explicitly covered, and ONDEWO announces three to four weeks to teach the engine a new language or dialect. For complex topics with highly specialised vocabulary, such as product names or support ticket categories, models are tailored by transfer learning.
Two recognition modes are offered. A real-time mode transcribes an audio stream and continuously updates its hypothesis of the transcription as speech is heard, and a batch mode processes WAV files. Both telephone audio at 8 kHz and mobile, laptop or tablet audio at 16 kHz are handled, while models for radio messages are described as under development. A single installation is said to run hundreds of parallel transcriptions, which is the scale a contact centre needs.
Integration goes through ONDEWO client libraries for Python, Nodejs, Angular and JavaScript, or through gRPC remote procedure calls, which the site itself spells GRRPC in both its English and German versions. Deployment is either on-premise, with a standard Docker environment on any operating system, or in a cloud provided by ONDEWO; all features are available in both versions, and the on-premise route is positioned as allowing a higher degree of data protection and control. ONDEWO develops its models on an NVIDIA DGX A100 supercomputer.
What it does
- Transcribe human voice, captured as audio, automatically into text.
- Transcribe a live audio stream in real time, with the transcription hypothesis refined as listening continues.
- Transcribe WAV audio files in batch.
- Train a model on a highly specialised business vocabulary through transfer learning.
- Run hundreds of parallel transcriptions from a single installation.
- Transcribe telephone audio at 8 kHz as well as mobile, laptop and tablet audio at 16 kHz.
- Drive applications, devices and processes by voice.
When to use ONDEWO Speech-to-Text / When not to
A quick filter to help you decide if ONDEWO Speech-to-Text is the right fit.
When to use ONDEWO Speech-to-Text
- Enterprises handling high telephone volumes - contact centres, support desks and helpdesks - that need hundreds of parallel transcriptions from a single installation.
- Organisations bound by data sovereignty or confidentiality requirements, which need the engine to run inside their own infrastructure rather than in a third-party cloud.
- Teams working in domains with highly specialised vocabulary, such as product names or support ticket categories, where general-purpose speech models break down.
- Regulated and public-sector operators in the industries ONDEWO addresses: telecoms, financial services, insurance, healthcare, energy, logistics, government and emergency call centres, including named projects with Frequentis and the FFG KIRAS project with Johanniter Austria.
- Technical teams comfortable deploying a Docker container and integrating a gRPC API through the ONDEWO client libraries.
When not to use ONDEWO Speech-to-Text
- Individuals and small teams looking for self-service: there is no sign-up, no trial and no published price, and every route to the product runs through a sales expert.
- Anyone expecting a mobile or consumer-facing application: S2T is a server platform driven entirely by API, with no App Store or Google Play listing.
- Buyers who need ready-made third-party connectors such as Zapier, Slack or Teams: none is announced, so every integration has to be coded.
- Organisations that must budget before they talk to a vendor: with no pricing page and no rate card, the cost of the platform cannot be estimated from public information.
- Projects that need, on day one, a language outside the four ONDEWO actually names - German, English, Spanish and French - since a new language or dialect is announced as a three-to-four-week training effort.
How to use ONDEWO Speech-to-Text
A typical end-to-end flow, from setup to results.
- Start with a sales conversation: there is no self-service sign-up, so the entry point is the expert contact form or a meeting request.
- Fill in the expert form, which asks for first name, last name, job title, company, email, phone, website and industry, or book a slot directly with the CEO through the HubSpot meetings link.
- Agree the deployment model with ONDEWO: on-premise on your own infrastructure, or the cloud option hosted and maintained by ONDEWO and billed on a usage basis.
- For on-premise, install the platform with a standard Docker deployment environment on your own IT infrastructure; all features are available either way.
- Choose an integration route: an ONDEWO client library for Python, Nodejs, Angular or JavaScript, or direct gRPC calls generated from the public protobuf definitions, which cover more than ten programming languages.
- Configure the client with the host, the port (6600, for example) and the gRPC certificate that secures the channel.
- Optionally enable Keycloak authentication, supplying the base URL, the realm, the public client id and a technical user.
- For real-time work, select audio stream transcription, click the microphone and speak; the text is transcribed as you go and can then be copied.
- For batch work, send a WAV file to the platform.
- If a business-specific vocabulary, language or dialect is needed, commission a custom model, for which ONDEWO announces three to four weeks, and raise any incident through the Bitbucket customer issue tracker, since no support email address is published.
Pros & Cons
Pros
- Genuine on-premise deployment on the customer's own infrastructure, a sovereignty and confidentiality argument that remains rare on this market.
- Custom models built by transfer learning for complex business vocabularies, with a word error rate claimed below 10%.
- Performance figures are published rather than merely implied: 85-95% accuracy and transcription 40 to 60 times faster than cloud providers.
- Real-time and batch modes cover both live telephone streams and stored files, including the low-bandwidth 8 kHz telephone channel that many general-purpose engines handle poorly.
- The gRPC API is documented by public protobuf definitions on GitHub, from which a client can be generated in more than ten programming languages.
- The official client library is actively maintained: ondewo-s2t-client reached version 7.4.2 across 30 releases, the most recent published on 21/08/2026.
- European publisher subject to the GDPR, with verifiable public references such as Frequentis and FFG research projects, and figures that match exactly between the English and German versions of the site.
Cons
- No pricing is public: there is no pricing page among the 65 pages of the site, no amount and no rate card, so the platform cannot be budgeted without contacting sales.
- No free trial and no free plan is announced anywhere.
- The published terms of use govern an earlier, discontinued product, a services marketplace, and not this platform, which is therefore sold with no published contractual terms.
- The privacy policy states explicitly that it covers the website only, so the processing of audio and transcripts by the platform is entirely undocumented.
- No security certification is published, neither ISO 27001 nor SOC 2 nor TISAX, and the list of processors is supplied only on request.
- No position is declared on training from customer data, no opt-out mechanism is documented, and no hosting country is stated for the cloud option.
- Only four of the ten advertised languages are named, no support email exists since support runs through a Bitbucket issue tracker, the company timeline stops at 2020, and a typo, GRRPC for gRPC, persists in both language versions of the product page.
Pricing & Plans
No pricing is published. There is no pricing page on the site, and no amount, plan or rate card appears anywhere across its 65 pages; the product is acquired exclusively through contact with a sales expert. Neither a free plan nor a free trial is announced. The only billing arrangement stated concerns the hosted option: the cloud option is billed on a usage basis, a wording that matches word for word in the German version of the same page. No commercial terms are published for the on-premise licence. Readers should note that the only monetary amounts present on the site are contractual penalties set out in the terms of use of an earlier, discontinued product, and are in no sense prices for this platform.
Data, GDPR & hosting
A consolidated view of how ONDEWO Speech-to-Text handles your data.
GDPR overview
ONDEWO GmbH is established in the EU (Vienna, Austria), so no Article 27 representative is required and none is named. The privacy policy cites the Datenschutzgesetz 2018, the DSGVO and the Telekommunikationsgesetz 2003, and lists rights of information, rectification, erasure, restriction, portability, withdrawal of consent and objection. The competent supervisory authority is named: the Austrian Data Protection Authority, Wickenburggasse 8, 1080 Vienna, dsb@dsb.gv.at. ONDEWO states that processing agreements (AVV) under Article 28 GDPR are concluded with its processors, but that list is not published and is only supplied on request to office@ondewo.com. No data protection officer is named; rights requests go to the same address. A consent age of 14 is stated, again for the website. No certification such as ISO 27001, SOC 2 or TISAX is claimed anywhere, and the policy itself covers only the website, not the S2T platform.
Who owns the data?
No published contract describes who owns the audio, the transcripts or the custom models the platform processes. The privacy policy states in as many words that it applies to the website ondewo.com, and the terms of use linked from the footer govern an earlier, discontinued ONDEWO product, so neither document speaks for the S2T platform. For website data only, ONDEWO says you retain control over the personal data you provide and that it will not be passed on to third parties beyond its own processors. In an on-premise deployment the software installs on the customer's own IT infrastructure, so the audio never leaves that perimeter, but that is a technical consequence of the architecture, not a contractual guarantee.
Reuse rights
Nothing states what a customer may do with its own transcripts, because no contractual text covers the platform: the terms of use linked from every footer govern a discontinued marketplace product, and the privacy policy declares that it applies to the website only. The documented uses therefore concern website visits. ONDEWO invokes consent, contract performance, legal obligation and legitimate interest under Article 6 of the GDPR, records automatic server logs (referring page, IP address, date and time, volume transferred, success or failure, operating system and browser), and runs Google Analytics with transfer and processing on servers in the United States, IP anonymisation being mentioned; LinkedIn and Facebook plugins forward no data on a simple visit. On the platform side, the only training described is transfer learning that tailors a model to the customer's own use case, never presented as feeding ONDEWO's general models. The vendor declares no position on using customer data for training and documents no opt-out mechanism.
Data retention & training
Hosting summary
Two deployment modes exist. On-premise, the platform installs with a standard Docker deployment environment on the customer's own IT infrastructure, on any operating system; ONDEWO presents this route as allowing a higher degree of data protection and control, which in practice means the audio never leaves the customer's own perimeter. The alternative is a cloud provided by ONDEWO, which requires no maintenance from the customer and is billed on a usage basis. All features are available in both versions. No hosting country and no hosting region is declared for the cloud option. One point of confusion is worth removing: the ondewo.com showcase site resolves to 34.89.223.2, an address operated by Google LLC in Frankfurt, Germany, but that is where the marketing site is served and it says nothing about where the platform would process customer audio. Separately, and again for the website only, Google Analytics transfers visit data to servers in the United States. Anyone with a jurisdictional requirement should either take the on-premise route or obtain the cloud hosting location in writing.
Things to keep in mind
Risks and trade-offs to weigh before adopting ONDEWO Speech-to-Text.
- The terms of use linked from every page footer, including the S2T product page, govern an earlier ONDEWO services marketplace, complete with members, registrations, offers, orders, a review system and Facebook Messenger. Their own scope clause even limits them to ONDEWO's German-language offers. Until terms are supplied to you directly, treat this platform as contractually undocumented.
- The privacy policy declares that it applies to the website ondewo.com. Nothing published states how audio, transcripts or custom models are handled, so the retention periods and legal bases you can read online do not apply to the platform.
- No opt-out from training is documented and no position is declared on the use of customer data. Do not assume either way without a written commitment in your contract.
- No hosting country or region is stated for the cloud option. The Frankfurt address behind the website's IP shows where the marketing site is served, not where your audio would be processed.
- No security certification is published, neither ISO 27001 nor SOC 2 nor TISAX, and the list of processors is only supplied on request, so the accuracy, availability and security claims rest on the vendor's own word.
- Accuracy figures are vendor claims measured on undisclosed data, and human level word error rate is not zero. In a contact centre, an emergency dispatch room or a clinical documentation workflow, treating an automatic transcript as a verbatim record, or letting it replace actually listening to the person speaking, is where real harm occurs. Benchmark on your own audio, especially for dialects, non-native speakers and 8 kHz telephone channels, and keep a human in the loop on anything consequential.
- Several signals point to a thin information layer around a maintained product: only four of the ten advertised languages are named, support runs through a Bitbucket issue tracker with no support email, the company timeline stops in 2020, the incorporation date is published only as a year with no day or month, and the product page has carried the typo GRRPC for gRPC in both its English and German versions.
Setup & Integrations
Technical difficulty
Moderate to high, and never self-service: with no sign-up, every deployment starts with a sales conversation. On-premise installation is announced as straightforward, a standard Docker environment on any operating system, but integration requires development, either against an ONDEWO client library or directly in gRPC, with a host, a port and a gRPC certificate to configure and optional Keycloak authentication. Custom models for a new language or dialect are announced at three to four weeks. Expect to need a technical team covering both infrastructure and development; this is not a tool an end user sets up alone.
Deployment
Supported languages
Behind ONDEWO Speech-to-Text
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
What does ONDEWO Speech-to-Text (S2T) Platform do?
How accurate is the transcription?
Which languages are supported?
Does it work in real time, or on recorded files?
Which audio channels are handled?
Can the platform be hosted in-house?
Is there an API, and how is the platform integrated?
How much does it cost?
Is there a free trial or a free plan?
Who publishes the tool, and how is support handled?
Should you pick ONDEWO Speech-to-Text?
ONDEWO Speech-to-Text (S2T) Platform is a technically credible product. Its publisher puts numbers on the table rather than adjectives - 85-95% accuracy, word error rates below 10% on customised models, transcription 40 to 60 times faster than cloud providers - the gRPC interface definitions are public on GitHub, and the official client library is actively maintained, its latest release dating from 21/08/2026. The positioning is equally clear: data sovereignty through a real on-premise Docker deployment on the customer's own infrastructure, and models tailored by transfer learning for the business vocabularies that defeat general-purpose engines.
The main reservation is commercial opacity. Nothing is priced: no pricing page among the 65 pages of the site, no amount, no plan grid, no free trial and no free plan. The single billing statement available is that the cloud option is billed on a usage basis. Every route to the product goes through a sales expert.
The second reservation is documentary. The terms of use linked from every footer govern an earlier ONDEWO product, a German-language services marketplace, and not this platform, which is therefore sold without published contractual terms. The privacy policy states that it applies to the website ondewo.com, so nothing published describes what happens to the audio and transcripts the platform processes. Even the company's incorporation date is published only as a year, without a day or a month, and no security certification is claimed.
The audience is unambiguous: enterprises with a technical team able to run a Docker container and code against a gRPC API, not individual users. The showcase site is dated, its company timeline stopping in 2020, while the product itself is demonstrably still maintained. Expect to obtain in writing, during the sales conversation, everything the site does not publish.
- Choosing a selection results in a full page refresh.
- Opens in a new window.