Adobe Podcast.

Adobe Podcast Product

I lead product for Adobe Podcast—developing AI tools that enhance spoken audio. I work closely with research and engineering to turn audio models into tools like Enhance Speech—improving clarity by reducing noise—and Studio, which enables multi-language speech-to-text, making it easier to edit speech like a document.

Enhance Speech Product Feature

I partnered with Adobe research, design, and engineering to productize generative audio models for stem separation and speech enhancement. Enhance Speech helps users clean up recordings for interviews, podcasts, and more—refining dialogue, reducing noise, and delivering studio-quality audio without professional equipment or post-production tools.

Speech to Text Product Feature

By leveraging generative audio models, I helped build multi-language speech-to-text transcription in Studio, allowing users to seamlessly record and transcribe audio. Users can edit transcripts like a document, making it easier than ever to edit audio without complex tools or manual waveform editing.

Home Studio

Travel Influencer

Toggle between the original and enhanced track to hear the difference

Enhance Speech

Enhance Speech removes background noise and elevates audio quality, making studio-like sound accessible to anyone. It's been a firsthand look at how AI is transforming audio creation. With v2, Enhance Speech delivers clearer, more natural dialogue even in noisy environments or poor acoustics. The updated strength slider gives users control over speech clarity and ambient noise balance, allowing for a customizable mix. It removes background noise, reverb, and music while keeping voices crisp and improves quiet or distant recordings without amplifying noise.

As part of the development process, I built custom tools using Claude and Cursor to accelerate testing and refine requirements. One tool (Audio Mixing) helped automate A/B comparisons by generating an audio file that alternated between original and enhanced segments every five seconds, making it easier to assess model quality before release.

Another tool (Stem Mixing) enabled dynamic stem mixing, allowing me to experiment with different mixing techniques for enhanced speech, background noise, and reverb. This helped evaluate potential remixing workflows without requiring upfront engineering investment, providing a fast, iterative way to explore more advanced user experiences.

Try the audio mixing tool Claude

Generate an audio file that switches between the original and enhanced versions every five seconds, making it easy to compare quality differences at a glance.

Try the stem mixing tool Claude

Upload and remix individual audio stems in real-time to preview how speech, background, and reverb blend together before finalizing the mix. Save each mix as a preset for easy resue.

Speech to Text

Speech-to-text, powered by generative AI, converts recorded audio into editable transcripts, simplifying workflows like podcast editing, captions, and audiogram creation for social media. In developing this feature, I focused on refining transcription quality—assessing punctuation, handling of complex terms, abbreviations, proper nouns, and natural speech elements like laughter and pauses. To streamline evaluation, I built a tool (Transcript Comparison) using Claude that compares transcripts across model versions, automatically highlighting filler words, repetitions, numeric data, and abbreviations to accelerate review and refinement.

Try the transcript review tool Claude

Upload and compare transcripts across model versions, automatically highlighting filler words, repetitions, numeric data, and abbreviations to accelerate review and refinement.

Adobe Podcast

podcast.adobe.com

AI-powered audio tools that elevate your voice. Create high-quality podcasts and voiceovers that sound professional with Adobe Podcast.

Podcast →