翻訳待ち：I've dictated over 120,000 words with my voice - these are my 3 favorite tools (and one is free)

AI サービスが一時的に利用できないため、復旧後に翻訳を補完します。

Follow ZDNET: Add us as a preferred source on Google. ZDNET's key takeaways Fast self-correction matters more than raw accuracy alone.Wispr Flow led on corrections, vocabulary, and reliability.Free, local FluidVoice nearly matched the paid winner.Since February, I have dictated 120,896 words. In the last three months alone, I have dictated 50,234 words. I've done this across 2,314 individual microphone recording sessions, averaging about 24 words per dictation sequence. With improvements in AI, dictation quality, speed, and accuracy have come a long way. Over the years, I've tried to work with speech recognition many times, but it hasn't been very successful until quite recently. A typical article is roughly a thousand words. So if you look at it that way, I have dictated the equivalent of roughly 120 articles since February. Also: 7 surprisingly useful ways to use ChatGPT's voice mode, from a former skepticThat's not to say I don't type. I type a lot, but it's clear that I also use dictation a lot. I've adopted voice dictation as a primary input modality for two key reasons. First, it helps to protect my wrist, which tends to have carpal tunnel symptoms. By dictating, I'm using my wrist a little bit less. In this article, I'll show you the three primary contenders that I've looked at and spent time with this year. I'm also going to go over a number of also-rans and honorable mentions, just to give you an idea of the dictation tools that are out there and how they differ. What makes a dictation tool worth using When it comes to dictation products, there are a variety of factors to consider. Price is certainly one of them. But there's also the question of whether or not you want your words sent up to the cloud to be processed or whether you want to keep them private and local. Accuracy is another factor to consider. Back in the day, it used to be much more of a problem. Fortunately, the voice models have achieved a level of overall quality that allows basic English words to be transcribed quite accurately. Some of the dictation products support other languages, but since I work only in English (at least for human languages), that's what I'm covering in this article. You also need to be able to correct your words as you're dictating. Some AIs go back and clean up mistakes you've made and provide you with clean text in whatever it is you're dictating into. Also: The best text-to-speech tools of 2026: Expert testedI am dictating most of this article just as a proof of concept to make the point that an article can be dictated. But that's not the norm for me. Despite all the dictation I do, I don't usually dictate my articles. I sometimes dictate a few lines of an article, but mostly I type them out. One of the reasons I don't use voice dictation for my articles is that voice dictation isn't particularly good at formatting. It's quite good at just laying down text, but if you want to arrange something, create a table, create bullets in a certain format, have indented quotes, or add any of the physical artifacts of an article, you have to go back and do that by hand after you dictate. Because of the regular need to do editing and superfine word crafting, I don't find voice dictation to be nearly as effective as hand typing on a keyboard like a caveman. Where voice dictation fits into my workflow On the other hand, I use voice dictation a lot when working with Claude Code or OpenAI's Codex to vibe code any of the products I'm working on. I use it a lot in Slack and Google Chat when talking to my editors and some of my project partners. Also: I built an iOS app in just two days with just my voice - and it was electrifying I use voice dictation quite a lot when replying to email messages. I use it a little bit less when composing new messages, but I use it there sometimes as well. I use it a lot when taking notes. I also use it sometimes for search phrases in Google or prompts that I give to ChatGPT, Gemini, or Claude Code. (Disclosure: Ziff Davis, ZDNET's parent company, filed an April 2025 lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.)Also: I built two apps with just my voice and a mouse - are IDEs already obsolete? I find that voice dictation is sometimes a little bit less precise and a little bit less disciplined than writing each word with the keyboard. But I can dictate at an average of 118 words per minute, while my typing is usually in the 70- to 80-word-per-minute range. When I'm working on an article, I sweat each sentence, choosing the words and structure with great care. Typing lends itself to that kind of attention to detail. But when I'm vibe coding, for example, and I'm discussing how I want a feature to be instantiated, how I want something to behave, or a bug that I've noticed, speaking is considerably faster and also considerably more gentle on my hands than typing it into the computer. The features that matter most I find two features to be mission-critical. The first is a customizable dictionary, so that when I say something like ZDNET, the dictation product understands how I want it spelled and presented. The second is an on-the-fly correction capability, so that when I say something and then correct myself and re-say it, the version that lands in whatever I'm dictating into contains those corrections. One oddly missing feature in all of the voice dictation products is the ability to go back and selectively edit already pasted-in words. Nothing can do this except the native Mac Voice Control or Windows Voice Access that comes with the operating systems themselves. Unfortunately, those two capabilities often conflict with the voice dictation products in inconvenient ways, so they don't combine well as integrated solutions. Also: I'm an AI tools expert, and these are the 4 I pay for now (plus 2 I'm eyeing)Almost all the dictation products are initiated by a hotkey. I bind the dictation hotkey to a button on my mouse so that when I tap the button, the dictation starts or stops. This allows me to dictate regardless of what application or web page I'm in at the moment. It means I can do computer input even if my keyboard isn't in front of me. In addition to the simple ability to listen to speech and turn that into typed words, most of the dictation products come with a dizzying array of configuration options, specialty features, and computer command capabilities. For example, some specialize in post-processing dictated speech with an AI for specific applications, like auto-formatting email messages. Others have features like transcribing an audio file. And nearly all of them give you some choice over what language model is being used for the dictation processing. I'm not going to spend much time on those bells and whistles in this article. Instead, I'm going to focus on the most important aspect, which is how well the tools turn speech into written text. To that end, I will be spotlighting three products: , , and That's next. Wispr Flow: Best overall At $144 a year, or $15 a month, is certainly not cheap, but I would argue it's actually worth it. Despite trying almost all the other products, this is the one I keep coming back to and have used more than any other. Wispr Flow is the only one of our top three available for Mac, Windows, iOS, and Android. It is not, however, available on Linux, although the company has a waitlist for Linux users. Wispr Flow's standout feature, at least in terms of my usage, is its in-flight self-correction. As you're dictating, you can correct yourself, and it updates what's being transcribed. Once you get used to this feature, you really don't want to go back, especially if you're doing a large amount of dictation like I do. It means that the text you produce is, more often than not, usable because if you misspeak, you can fairly easily correct it as you're speaking and end up with a decent result. None of the other models that I tested were able to do this as smoothly. Some couldn't do it at all. FluidVoice has come close, but I would say that Wispr Flow made accurate corrections eight out of 10 times, and FluidVoice made accurate corrections maybe four out of 10 times. For in-flight self-correction, that's measurable when you're doing a lot of work. Also: I tested 3 text-to-speech AI models to see which is best - hear my resultsI also found that Wispr Flow's dictionary is reliable and effective. What I mean by that is that once I've trained it on an incorrectly spelled word or incorrectly interpreted word, I almost never have to go back and correct it again. Once I trained it on the word ZDNET, for example, Wispr Flow reliably gets it correct just about 100% of the time. That's also the case with my library of 90 or so other words that I regularly correct. Wispr Flow has two dictionary options: It allows you to feed it individual words like Gewirtz, and it allows you to feed it misspellings or misinterpretations and then the corrected word. For example, it regularly had trouble with the word Claude, which it would represent as "call it." I set up a dictionary definition for "call it code" that converted to Claude Code, and I've never had a problem since. Once in a while, Wispr Flow misses the insertion of a chunk of text into the destination location. For example, I might dictate a paragraph that I want to go into Notes, and it never winds up there. Wispr Flow keeps a history of dictation in its app. If it misses insertion, I can open it up in the app, copy from the history, and paste it in. I don't ever actually lose any of my dictation, even if it doesn't always arrive on target the first time out (which is a fairly rare occurrence). Beyond price, my biggest concern about Wispr Flow is that it's a cloud-only model, meaning that all of your voice snippets are sent to the cloud for transcription. Despite the similarity in names, Wispr Flow is not based on OpenAI's open-source Whisper speech recognition technology. Wispr Flow appears to be its own model or based on a stack of a variety of model providers. The company does not disclose the exact model used. Also: I tested ChatGPT's Live Voice upgrade, and it almost felt human - how to try itWispr Flow offers a number of data and privacy options in its settings, including a privacy mode, the option to turn private cloud sync on and off, and local data storage. However, what the options are called in the UI and what the options actually do are different. Privacy mode isn't really what you would think. It's not that it doesn't look at any of your phrases. It's that when turned on, it will not send any of your data to be used for training the AI. Private Cloud Sync, when turned off, does not mean that the data is not sent up to the cloud. It means that it's not stored in the cloud to sync to other devices. It is still sent up to the cloud for transcription, but Wispr Flow then deletes the data immediately after transcription. The local data storage option does not control whether data is stored locally or in the cloud, but instead controls factors like whether or not Wispr Flow will auto-delete local data every 24 hours or never store any data locally, meaning, for example, that the dictation history would not be available to you. Also: I used Gmail's AI tool to do hours of work for me in 10 minutes - with 3 promptsIf you have data control policy concerns, confidentiality concerns, disclosure restrictions, or any other legal reason you don't want your data up in the cloud, you might want to avoid Wispr Flow. I have found, for basic productivity, that Wispr Flow has become my most actively used voice dictation product. I have been cycling through a bunch of them to try to find one that I could live with as a daily driver. So far, that's Wispr Flow, and that's why it's my top recommendation. Wispr provided me with a Pro account to use for a year for evaluation, but there's a very good chance th [truncated for AI cost control]