VoiceInput vs Wispr Flow

Wispr Flow is a polished English-first dictation tool with strong AI rewrite features. VoiceInput targets Chinese and mixed-language typing speed, ships sub-1.5-second latency, and turns every utterance into a searchable memory you can revisit a month later.

One-line verdict

If you write English emails all day and want AI to clean up your tone on the fly, Wispr Flow's rewrite features are excellent. If you type Chinese or mixed Chinese-English, want under-1.5-second latency, and want everything you said to be retrievable later — VoiceInput is built for that.

Side-by-side comparison

Where Wispr Flow wins

• English tone rewrite. If your job is writing English emails, Slack messages, and tweets, Wispr Flow's rewrite presets are mature and useful. VoiceInput's AI tidy intentionally won't rewrite your tone — only fix homophones, drop fillers, and add punctuation.
• Brand polish. Wispr Flow has invested heavily in onboarding and visual design. The product feels finished from the first second.
• Native English ASR. Their cloud model is tuned for English and performs well on accented speech.

Where VoiceInput wins

• Speed on Chinese. Volcengine streaming ASR + local typography engine = sub-1.5s on Mandarin and mixed CJK-English. Wispr Flow is 3-10s on the same input — that's the difference between dictation feeling natural and feeling like you're waiting on it.
• Memory layer. Every voice line auto-archives. Search "what did I say about onboarding last month?" — Wispr Flow drops the text and forgets.
• Persona reviews. Same line re-read by Boss / Coach / Therapist / Editor. Weekly Big5 sketch derived from how you actually talked. This is a different product category, not just dictation.
• Local-first option. 100% offline mode is real, not theoretical — three local ASR engines plus a local typography engine. Wispr Flow is cloud-only.
• Free forever (Local + BYOK). No card. No rate limit on the local tier. Wispr Flow's free tier caps usage and pushes Pro hard.
• BYOK. Plug in DeepSeek / Kimi / OpenAI / your own server. Pay model providers directly (~$0-2/month for typical use). Wispr Flow doesn't expose this.

The tone-rewrite question

Wispr Flow's rewrite features (make this email more formal, make this tweet shorter) are the product's signature. VoiceInput intentionally doesn't do this:

• The AI tidy prompt is constrained to three jobs: fix homophones, drop fillers, add punctuation. Confidence below 0.5 keeps the original.
• Double-tap right Option to bypass AI cleanup entirely — get the raw ASR output.
• Tone shaping happens in the memory layer instead — 7 personas re-read what you said and offer different angles. You read it later, not at the moment of dictation.

This is a deliberate split. If you want voice → polished output instantly, Wispr Flow wins. If you want voice → fast clean text + retrievable memory, VoiceInput wins.

Speed: why 1.4 seconds matters

Latency under 1.5 seconds is the threshold where dictation stops feeling like a tool and starts feeling like typing. Above 3 seconds, you wait — and the flow breaks.

VoiceInput's pipeline:

1. Persistent ASR connection — no handshake per utterance.
2. Streaming partial transcripts — text starts arriving before you stop talking.
3. Local typography engine handles formatting in <5ms (no LLM call).
4. Optional AI tidy fires only after pause detection — never blocks the first injection.

Wispr Flow's pipeline batches audio in chunks before sending — built for transcription accuracy on English, not real-time CJK throughput.

Memory layer: a different product category

Most dictation apps end at "speech became text." VoiceInput treats every utterance as data worth keeping:

• Tool (SPEAK). Hold to talk, text lands at the cursor. Same job Wispr Flow does.
• Data (RECALL). Every line archives locally with source app, time, tags. Full-text search across months. Export to Markdown / JSON / CSV.
• Memory (REFLECT). 7 personas re-read your week. AI picks 3-5 quotes worth echoing. Weekly Big5 snapshot derived from real talk — patterns you wouldn't see yourself.

Privacy

• VoiceInput. Audio + text + history all live on your Mac. Cloud ASR streams audio directly to Volcengine — no relay server. Only metric leaving the device is recording length in seconds, for the global pulse counter (toggle off in Settings). API keys live in macOS Keychain, never on our servers.
• Wispr Flow. Cloud-first SaaS. Standard retention and processing policies apply. No local-only mode at the time of writing.

Who should pick which

FAQ