# Speech enhancement

> Speech enhancement reduces unwanted noise or artifacts in recorded speech; AI-processed audio still needs comparison with the original for intelligibility.

Speech enhancement reduces unwanted noise or artifacts in recorded speech; AI-processed audio still needs comparison with the original for intelligibility.

**Speech enhancement processes existing speech to reduce unwanted noise or artifacts and improve clarity.** Adobe describes its AI Enhance Speech feature in those terms and separates it from Mic Check, which addresses setup before recording. [Adobe's introduction](https://podcast.adobe.com/guides/what-is-enhance-speech) explains the distinction.

The result depends on the input: Adobe notes that audibility and background noise affect its output. [Its Podcast FAQ](https://helpx.adobe.com/podcast/adobe-podcast-faq.html) describes that limitation. A processed recording therefore remains material to compare with the original.

In the fictional neighborhood interview, listener J. Chen records “the west bench” in the original and “the wet bench” in the processed copy at the same interval. A quieter background would not settle that changed word. The listener keeps the original for that window. Another interval has quieter traffic and retained words, so the listener accepts that processed portion separately.

The [speech-cleanup workflow](/guides/voice/review-ai-speech-cleanup) records paired observations, time bounds, decisions, and pending work. A processing surface such as [Adobe Podcast](/tools/adobe-podcast) can provide the variant being reviewed, while the listener decides whether its speech should be used.

Transcription turns speech into text, and text-to-speech creates spoken output from text. Enhancement works on an existing speech recording. In this review workflow, the [Speech Cleanup Reviewer](/agents/marketing/speech-cleanup-reviewer) organizes supplied observations; it cannot claim an audition merely because it can read file names or notes.

Keep the original unchanged and mark unreviewed intervals pending. A complete row shows who compared the two versions and why they chose one, but a structural checker cannot verify that listening occurred. The acceptance decision still belongs to the listener before an interval or final master is used.

---

_Source: https://agentscamp.com/glossary/speech-enhancement — Term on AgentsCamp._