What is the difference between speech input and read-aloud plugins?

Speech input and read-aloud plugins are different: speech input turns voice into text in the input box, while read-aloud turns replies into speech. They can be installed together.

Does read-aloud need a separate API key, and does it cost extra?

Read-aloud may need a separate API key depending on the provider. Some DeepSeek Harness plugins use a third-party TTS service with its own key and cost.

Speech recognition is inaccurate. Is it the plugin or the microphone?

Inaccurate speech recognition usually starts with microphone permissions and the input device. If those are fine, check noise, language setting, and network.

What is a proactive follow-up voice plugin, and does it interrupt?

A proactive follow-up voice plugin speaks prompts based on rules, and interruption depends on configuration. Check frequency and mute options before enabling.

Voice plugins: input, read-aloud, TTS and follows

Voice plugins are not one category but four directions: speech input, read-aloud, TTS synthesis, and proactive follow-ups. Knowing the direction avoids installing the wrong plugin expecting the opposite.

Four directions

Separate by direction first.

Type	Direction	Typical use	Cost
Speech input	Voice to text	Hands-busy input	Recognition service
Read aloud	Reply to speech	Listening to answers	TTS service
TTS synthesis	Text to audio file	Saving or sharing	TTS service
Proactive follow-up	Rule to speech	Reminders and prompts	Depends

Input and output are opposite directions. Input puts your voice into the box; read-aloud puts replies into your ears. They can coexist, but they solve different problems.

TTS synthesis is often part of read-aloud. It can also export audio separately; check whether the plugin provides a file output.

Proactive follow-up is rule-driven. It speaks on a schedule or condition rather than on your request.

Speech input

The most common use: talk instead of typing.

Confirm microphone permission for the browser or app.
Grant permission on first use; do not dismiss the prompt.
Choose the input device and language.
Speak one sentence, confirm the text, then send.
In noisy environments, use a headset or quieter place.

Permissions are the most common failure. Denied once in the browser, the plugin will not hear you until you re-grant.

Recognition quality depends on environment and language setting. Wrong language setting produces recognizable but wrong text; confirm the language first.

Read-aloud and TTS

Turning replies into speech; configuration centers on voice, speed, and trigger.

Choose auto-read or manual read.
Choose voice and speed; too fast hurts comprehension.
Confirm TTS provider and key, if required.
Turn off auto-read for long code blocks.
Mute or pause on demand.

Auto-read suits listening; manual suits when needed. Auto-read on every reply interrupts reading and consumes TTS quota.

Keys and costs vary by provider. Some use third-party TTS with keys and billing; check before installing (source).

Proactive follow-up

The most likely to be intrusive; rules decide.

Confirm what triggers it: time, task completion, or errors.
Set frequency; long intervals reduce interruption.
Provide mute and off switches.
Check voice and content; avoid private information.
Enable in quiet environments first.

Rules determine experience, not the voice. Good voice with bad timing is still noise.

Do not use it for reminders that need confirmation. Speech is one-way; use messages for confirmations.

Choice by scenario

Choose by what you need to convert.

Scenario	Type	Reason
Hands busy, need input	Speech input	Fastest entry
Listen while doing other work	Read aloud	Output to ears
Need audio files	TTS synthesis	File output
Reminders	Proactive follow-up	Scheduled prompts

You do not need all four. Start with the one covering your main scenario, then add.

Combine input and read-aloud for conversation. Talk and listen; the cost is recognition and TTS usage.

Failure and privacy

Voice adds hardware and privacy layers.

No microphone permission: check browser or system authorization.
Recognition inaccurate: check language, noise, and device.
No sound: check output device, volume, and TTS key.
Privacy: voice may upload to a recognition or TTS service; read agreements.
Cost: cloud TTS and recognition may bill by usage.

Check hardware before software. A muted microphone or wrong output device is the most common "broken" case.

Do not use voice plugins for sensitive content by default. Upload and retention depend on the provider.

Next steps

Choose direction, check permissions, watch privacy. For roleplay and persona plugins, read How to choose roleplay plugins. For plugin categories, read What DSH plugin categories exist. For desktop pets, read DeepSeek desktop pet setup.

Sources: dsh-speak-up plugin page, DeepSeek Harness docs, dsh CLI README