Official documentation
Speed of Sound FAQ
Antonio Zugaldia, Speed of Sound, 2026
Read the originalRead on October 20, 2026
- What it is
- The FAQ of Speed of Sound. It says that on-device inference is the default, that cloud providers are off unless configured, and that text polishing with an LLM is optional and disabled by default.
- How much weight it carries
- The project's own documentation, read on 2026-10-20. It states its defaults plainly, but the defaults can change with each release.
Facts taken from this source
- Speed of Sound's FAQ says text polishing is an optional step that sends the raw transcript to an LLM, and that it is disabled by default.
“Text polishing is an optional step that sends the raw transcript to an LLM for further improvement. It is disabled by default and, if enabled, it adds some latency depending on the provider and model you choose.” (FAQ, What does text polishing do, and do I need it?)
- Speed of Sound's FAQ says audio and transcriptions are never written to disk and are discarded as soon as they are processed. It gives no fixed number of days.
“Audio and transcriptions are never written to disk. They are discarded as soon as they are processed.” (FAQ: Where does the app store my models, settings, and logs?)
- Speed of Sound's FAQ says cloud connectivity is optional and can happen at transcription and at text polishing with an LLM.
“Cloud connectivity is optional and can happen at two points in the pipeline: transcription (ASR) and text polishing (LLM).” (FAQ, Does my audio or text get sent to the cloud?)
- Speed of Sound's FAQ says its LLM text polishing can take a custom vocabulary of names, companies or technical terms.
“It is most useful giving the LLM instructions and supplying a custom vocabulary of names, companies, or technical terms that speech recognition tends to mishear.” (FAQ, What does text polishing do, and do I need it?)
- Speed of Sound's FAQ answers no to whether audio or text is sent to the cloud, and says audio never leaves the machine unless cloud keys are added.
“No. Everything runs on your device by default. Audio and text never leave your machine unless you explicitly add API keys in the settings and enable a cloud-based provider.” (FAQ, Does my audio or text get sent to the cloud?)
- Speed of Sound's FAQ says on-device inference is the default, and that cloud providers are disabled unless configured.
“On-device inference is the default and works well for most users, and cloud providers are disabled unless you configure them.” (FAQ, Why do you support cloud providers at all?)