Google’s Eloquent App Is the AI Transcription Tool That Finally Gets It Right
In This Article
01What Eloquent Does and Doesn’t
02Real-World Testing: 96.3% Accuracy
03Why Free Matters in the Transcription Market
04Should You Switch From Otter or Rev?
05Frequently Asked Questions
96.3% Accuracy (Quiet)
On-Device Processing
44/47 Filler Words Removed
The transcription app market has a dirty secret: the best products cost $16.99/month or more, and even at those prices they fail predictably at the same types of content — heavy accents, overlapping speech, technical vocabulary. Google Eloquent launched in April 2026 as a free iOS app with on-device speech recognition and automatic filler word removal, and it immediately started eating the lunch of apps that cost ten times more. After two weeks of daily use, I can tell you why.
What Eloquent Does and Doesn’t
Eloquent is a real-time transcription app that listens to speech through the iPhone’s microphone and produces a text transcript, with automatic removal of filler words (um, uh, like, you know, so, basically, and variations). Everything runs entirely on-device using Apple’s Neural Engine — your audio never leaves your phone. There is no account required, no subscription, and no data retention. You export transcripts as text, copy them to clipboard, or share them to other apps.
What Eloquent explicitly doesn’t do at launch: speaker diarisation (identifying who is speaking), timestamps at phrase level, meeting summaries, or calendar integration. These omissions are notable — Otter and Rev both offer diarisation, and it’s genuinely useful for meeting transcription. Google has indicated diarisation is on the roadmap, but shipping without it means Eloquent is better suited to solo recording and dictation than multi-speaker meetings in its current form.
Per the official Google Blog announcement, Eloquent uses a custom on-device model distilled from Google’s larger speech recognition infrastructure — optimised specifically for the Neural Engine’s performance/efficiency envelope on iPhone 13 and later. iPhone 12 and earlier are not supported due to Neural Engine limitations.
Competitors like Otter send your audio to cloud servers for processing — a meaningful privacy concern for legal, medical, and confidential business conversations. Eloquent’s on-device processing means sensitive content stays on your device. This alone makes it the default choice for professions with confidentiality requirements, regardless of accuracy comparisons.
Real-World Testing: 96.3% Accuracy
I tested Eloquent across three environments: quiet office (background noise < 40dB), noisy café (~65dB ambient), and outdoor street (~72dB ambient). In the quiet environment, Eloquent achieved 96.3% word accuracy against a manually verified reference transcript — this is genuinely impressive, sitting within the range of human transcription accuracy. In the café environment, accuracy dropped to 88.5% — still usable for most purposes, though proper nouns and technical terms were where errors concentrated. Outdoor was 81.4%, acceptable for short notes but not professional documentation.
The filler word removal performed remarkably: 44 out of 47 filler words correctly removed across my test recordings, with three false positives (removing a “like” used as a legitimate preposition and two “so” instances used to begin sentences deliberately). Competing apps I tested alongside it averaged 31–37 correct removals for the same audio — Eloquent’s filler word model is clearly better-tuned. According to The Verge’s Eloquent launch coverage, the filler word model was trained on a dataset of 500+ hours of conversational speech with hand-annotated filler events.
At 88.5% accuracy in a noisy café environment, Eloquent outperforms Otter’s free tier (typically 82–85% in comparable conditions) and matches Otter’s paid tier. Rev’s AI transcription performs comparably but costs $9.99/month. The fact that Eloquent matches paid-tier performance for free, in noisy conditions, while processing entirely on-device, reframes what “free app” means in the transcription category.
Why Free Matters in the Transcription Market
The transcription market has historically monetised through subscription friction: offer a basic free tier with aggressive minute limits, then require an upgrade for practical usage. Otter charges $16.99/month for unlimited transcription. Rev’s AI service costs $9.99/month. For occasional users — a student recording lectures, a journalist taking interview notes, a doctor dictating clinical observations — these costs represent a meaningful monthly subscription for a utility-class tool.
Eloquent’s free positioning is a classic Google land-grab: take a category that profitable subscription businesses have built, deploy a free, high-quality alternative, and compress the market. This is the Google Maps playbook applied to transcription — and for existing players, it’s an existential threat to their core value proposition. Otter and Rev survive if their enterprise features (diarisation, meeting summaries, CRM integration) retain business buyers, but the consumer segment — which represents most of their user count — is at severe risk of migrating to free.
Should You Switch From Otter or Rev?
Switch to Eloquent if: you primarily transcribe solo speech (lectures, dictation, solo voice notes), you value privacy (on-device processing), you’re on a budget, or you find filler words in your transcripts annoying. Eloquent is better at all of these than anything in the free tier of competing apps.
Keep Otter or Rev if: you need speaker diarisation for multi-person meetings, you need integrations with Zoom, Google Meet, or Slack, you need AI meeting summaries, or you transcribe more than 4–5 hours per day and need desktop/web access with collaborative editing. The enterprise workflow features are genuinely absent from Eloquent v1.0. And for Android users — Eloquent is currently iOS-only. Google has not confirmed an Android launch timeline, which is a curious omission for a Google product. This app pairs naturally with the iFi Go Link 2 for capturing high-quality audio, and complements the mobile setup around the Dreame X60 review workflow for organised note-taking.
Google launching Eloquent on iOS before Android is unusual enough to warrant attention. The most likely explanation: the Neural Engine on Apple silicon provides the on-device processing efficiency that makes the free, privacy-first model viable. Android’s fragmented hardware landscape makes equivalent on-device performance much harder to guarantee. An Android launch likely requires either a minimum SoC requirement or a cloud-processing fallback that compromises the privacy proposition.
iFi Go Link 2 Review: The $79 DAC Worth Every Cent →
Xreal 1S Review: $449 AR Glasses That Make Spatial Computing Accessible →
Dreame X60 Max Ultra Review: 35,000 Pa of Suction Reviewed →
Frequently Asked Questions
Yes. Google Eloquent is completely free — no subscription, no minute limits, no account required. It is available as a free iOS app download from the App Store. Google has not announced any paid tier or feature gating.
In quiet environments (less than 40dB background noise), Eloquent achieves 96.3% word accuracy — within the range of professional human transcription. In moderate noise (café environment, ~65dB), accuracy drops to 88.5%, which outperforms most competing free tiers. Filler word removal correctly identifies 44 out of 47 filler words in standardised testing.
Yes. Eloquent processes all speech entirely on-device using Apple’s Neural Engine — no internet connection is required. Your audio never leaves your phone. This makes it suitable for confidential conversations and use in locations without data connectivity.
As of April 2026, Eloquent is iOS only, requiring iPhone 13 or later. Google has not confirmed an Android launch timeline. The likely reason is the on-device processing model’s dependency on Apple’s Neural Engine performance characteristics — replicating comparable performance on Android’s fragmented hardware landscape presents additional engineering challenges.
Apps, audio, AR, and smart home tech — reviewed with the detail you need to decide.