How we use AI at Spoken

At Spoken, we use artificial intelligence (AI) for one purpose only: automatic speech recognition (ASR), which we use to create raw text transcripts and subtitle drafts.

Our speech recognition models (such as Whisper or similar) run in our own controlled environment, locally. We do not send your audio or video files to public, so-called “black box” AI services.

The final output is always your choice: you can receive either a human-reviewed transcript or subtitles, or a pure AI transcript or subtitles delivered as-is.

How is AI used in our work?

When we use AI, the process is straightforward: an audio or video file is run through a speech recognition model, which produces a raw text output and, where needed, timecodes. Our professionals then edit, correct and finalise the text and subtitles if you have ordered a human-reviewed service.

Speech recognition is not perfect. It may omit words, break sentences in unusual places or mishear terms. When you order professional transcripts or subtitles from us, these errors are corrected manually, and we make sure the text matches the original speech and your brief.

We do not use AI as a separate machine translation service, nor do we let AI make decisions about people (for example scoring, profiling or selections). For us, AI is purely a technical aid for transcription and subtitling.

Human-reviewed or AI-only – what is the difference? 

In our human-reviewed service, every transcript and subtitle file delivered to you is checked by a human. We review the content, remove or correct any incorrect passages, harmonise terminology and format the text according to your instructions. Spoken is responsible for the quality of the final result.

If you prefer, you can also order AI-only transcripts or subtitles. In that case, you receive the raw speech recognition output without human review. We clearly indicate that these files are AI-generated drafts, which may contain errors and fragmented sentences.

Protection and use of client material

All client material – audio, video and text – is handled confidentially, including when AI is used.

Our speech recognition models run in our own technical environment, and we do not use public AI services that we cannot control. We do not use client material to train public AI models without explicit consent. Your material is processed only for the purpose we have agreed – typically transcription or subtitling. Access to the files is limited to the people at Spoken who work on your project.

We store and delete material in line with our internal data protection policy and any contractual requirements. More detailed descriptions can be found in our Privacy Policy and, where applicable, in project-specific data processing agreements.

AI and the EU AI Act

The EU AI Act categorises AI systems into different risk levels. The AI used at Spoken falls into a low-risk category: we use it only for automatic speech recognition, not for recruitment, credit decisions, surveillance or similar high-risk purposes.

In practice, this means our key obligations are:

• being transparent about where and how we use AI
• ensuring appropriate data protection and security
• keeping humans in charge of oversight and responsibility for the final result

We follow the implementation of the AI Act and maintain an internal record of the AI tools and practices we use, so we can update our processes when needed.

Questions

We want you to feel clear and confident about how your material is handled.

We can explain in more detail how your data is processed technically, adapt our process to your industry or organisation’s requirements, and help you choose the best option for your project:human-reviewed AI-assisted work, or a pure AI service.

For more information, please contact us via the details on our Contact us page.