ZDNET offers reliable coverage on technology topics and welcomes readers to add the site as a preferred source via Google.

ZDNET key takeaways

Quick self correction proves more valuable than sheer accuracy by itself.
Wispr Flow excelled in handling fixes vocabulary and consistent performance.
The free local option FluidVoice came very close to the leading paid choice.

Beginning in February I have spoken a total of 120896 words into my device. During just the past three months that figure reached 50234 words. These efforts occurred over 2314 separate microphone sessions with an average of roughly 24 words each time.

Advances in artificial intelligence have greatly improved dictation speed and precision in recent times. Earlier attempts to rely on speech recognition often fell short but recent progress has changed that situation.

An average article contains about one thousand words so the total spoken here equals roughly 120 full articles since February.

Typing still plays a major role in my routine yet dictation has become a frequent choice for two main reasons. It reduces strain on my wrists which sometimes show signs of carpal tunnel issues.

This review covers the three main tools I examined closely this year along with brief notes on other options to illustrate the range of available choices.

Key factors for selecting a dictation tool

Several elements influence the value of these products. Cost is important but so is the choice between cloud processing and keeping everything private on a local device.

Accuracy matters yet modern models now handle standard English text quite well. Support for additional languages exists in some cases but this discussion focuses on English only.

The ability to fix errors during dictation is essential. Certain systems automatically refine text as it appears in the target application.

I am speaking most of this piece as a demonstration yet that approach is not my usual method. Although dictation sees heavy use I rarely apply it to complete articles and instead type the majority by hand.

One limitation is formatting. These tools handle plain text effectively but tasks such as creating tables bullets indented sections or other layout elements require manual adjustments afterward.

Detailed editing and precise word selection remain easier with a keyboard than with voice input alone.

How dictation supports my daily tasks

Dictation sees frequent use when collaborating with coding assistants like Claude Code or OpenAI Codex. It also helps in messaging platforms such as Slack and Google Chat with colleagues and partners.

Email replies often rely on voice input while new messages use it less often. Note taking benefits as well and occasional search queries or prompts to various AI services also employ it.

Voice input can feel less exact than typing each word yet it reaches an average of 118 words per minute compared with my usual typing speed of 70 to 80 words per minute.

Article writing demands careful attention to every sentence so typing supports that focus better. In contrast discussing features or bugs during coding sessions benefits from the speed and reduced hand fatigue of speaking.

Essential capabilities to look for

Two features stand out as critical. A customizable dictionary ensures proper handling of terms such as ZDNET. On the fly correction allows rephrasing during speech so the final text reflects those changes.

Selective editing of already inserted text remains unavailable in most dedicated tools though native system features on Mac and Windows offer limited support that can interfere with other options.

Most products activate via a hotkey which I assign to a mouse button for quick access regardless of the current application. This setup supports input even when the keyboard is not nearby.

Beyond basic transcription many include extra settings such as post processing for specific uses or audio file handling along with model selection choices. The focus here stays on core speech to text performance.

The three products highlighted are Wispr Flow Superwhisper and FluidVoice.

Wispr Flow delivers the strongest results overall

Priced at 144 dollars yearly or 15 dollars monthly Wispr Flow costs more than alternatives yet the investment proves worthwhile based on repeated use. It remains the option I return to most often.

This tool works across Mac Windows iOS and Android though Linux support is still in development with a waitlist available.

Its strongest point is smooth in flight correction. Speaking a fix updates the output immediately and this reliability becomes especially useful during extended sessions.

Other tested options handled corrections less consistently. Wispr Flow succeeded about eight times out of ten while FluidVoice managed roughly four out of ten in similar tests.

The dictionary feature also performs reliably. Once trained on specific terms or corrections it maintains accuracy without further intervention. Examples include handling ZDNET and setting rules such as converting a misheard phrase into Claude Code.

Occasional insertion misses can occur but a built in history allows quick recovery without data loss.

All processing occurs in the cloud which raises privacy considerations for some users. The company does not specify the exact underlying models. Settings for privacy mode cloud sync and local storage exist yet their actual effects differ from initial expectations.

Privacy mode prevents data use for model training rather than blocking all access. Disabling cloud sync stops long term storage but transcription still happens remotely before immediate deletion.

Local storage controls mainly affect history retention and auto deletion rather than whether data leaves the device.

Users with strict data policies may prefer alternatives. For general productivity needs Wispr Flow has become the primary choice after comparing many options. A trial allows up to 2000 words of testing.