Key takeaways
- Speech carries tone and emotion that text loses, and AI now converts between them both ways.
- Computer speech synthesis dates to 1961, and neural voices arrived with WaveNet in 2016.
- Recognition problems usually come from microphones, noise, settings or connectivity.
- ChatGPT, Copilot, Word and PowerPoint all offer dictation or read-aloud features.
- Hyphenate speech-to-text before a noun, but not in the phrase convert speech to text.
Speech and text are two forms of language, and AI now converts between them: speech-to-text transcribes what people say, while text-to-speech reads text aloud with natural voices. Both are built into phones, Office apps and AI assistants. I built the OCR & Speech Workspace, which includes live speech-to-text, so this page covers speech versus text, TTS history, why recognition fails, TTS costs, ChatGPT and Copilot voices, Word and PowerPoint dictation, and hyphenation.
What is the difference between speech and text?
Speech is spoken language carried by sound, including tone, pace, emphasis and emotion. Text is written language, made of characters that can be read, searched and edited. Speech recognition converts speech into text, and speech synthesis converts text into speech.
Speech carries information text loses, such as sarcasm, hesitation and speaker identity, while text is easier to skim, store, translate and search. Spoken language also tends to be less structured, with fillers and restarts. For AI, the difference matters: transcripts lose prosody unless annotated, and synthesised speech must add natural rhythm. Modern multimodal models increasingly process audio directly, preserving more of speech's information than plain transcripts.
How long has text to speech existed?
Mechanical speaking machines date back to the eighteenth century, Bell Labs demonstrated the electronic Voder in 1939, and computer speech synthesis arrived in 1961, when an IBM computer at Bell Labs sang Daisy Bell.
Rule-based and concatenative systems followed, such as DECtalk in the 1980s, which became familiar through assistive devices. Concatenative synthesis stitched recorded speech units together, sounding clearer but less flexible. Screen readers made synthetic speech essential for blind and low-vision users. Neural text-to-speech, beginning with DeepMind's WaveNet in 2016, transformed quality, producing voices that are hard to distinguish from humans. Today's systems support many languages, emotional styles and voice cloning, raising both accessibility benefits and impersonation concerns.
Why does speech to text seem to be getting worse?
Speech recognition usually seems worse because of microphone problems, background noise, a changed device or app setting, network issues with cloud recognition, software updates that changed models, or speaking styles and vocabulary the system handles poorly.
Also asked as: why is speech to text getting worse
Check microphone permissions and input selection, clean or replace the microphone, and test in a quiet room. Updates sometimes switch between on-device and cloud recognition, which differ in accuracy. Accents, mixed languages and technical terms reduce accuracy, so use apps that support custom vocabulary or your language. Live transcription tools, including the speech-to-text in my OCR & Speech Workspace, depend heavily on clean audio input. Modern models such as Whisper are far more accurate than older systems, so persistent problems usually have a local cause.
Why doesn't your text to speech work?
Common causes include muted or wrong audio output, missing voice data or language packs, disabled app permissions, an unsupported language, browser settings blocking speech, or no internet connection for cloud voices.
Also asked as: why doesn't my text to speech work
Check volume and the selected output device first. On Windows, open Settings, Time and language, Speech, and install voices; on Android, check the text-to-speech engine settings and voice data; on iPhone and Mac, check Spoken Content under Accessibility. In browsers, try another browser or disable extensions that block audio. For apps using cloud voices, confirm connectivity and account limits. Restarting the app or device fixes many temporary failures, and reinstalling voices fixes corrupted voice data.
Why is text to speech so expensive?
High-quality neural voices require significant computing power to generate audio, licensed or recorded voice talent, and ongoing model development. Providers usually charge per character or per minute, so long content and premium voices add up quickly.
Voice cloning, emotional control and commercial usage rights cost more. Cheaper options exist: built-in operating system voices are free, cloud providers offer standard voices at lower rates than premium ones, and open-source models such as Piper and Kokoro run locally at no per-use cost. For accessibility features, built-in voices are often good enough. For large projects, compare quality and licensing terms carefully, and cache generated audio rather than regenerating identical content repeatedly.
Can ChatGPT generate text to speech?
Yes. ChatGPT can read responses aloud and hold spoken conversations through voice mode. Developers can generate speech files with OpenAI's text-to-speech API models, choosing voices and adjusting style with instructions.
In the ChatGPT apps, a read-aloud option plays any response, and voice mode allows natural back-and-forth conversation. Through the API, you send text and receive audio in formats such as MP3, useful for narration, accessibility features and voice assistants. Usage policies require disclosing that voices are AI-generated where relevant and prohibit impersonating real people. For long-form narration, compare voices from several providers, since quality varies by language and some voices suit certain content better.
Can Copilot generate text to speech?
Microsoft Copilot can speak responses in voice conversations, and Microsoft Edge's Read Aloud reads web pages and PDFs. For developers, Azure AI Speech provides neural text-to-speech voices in many languages through an API.
Microsoft 365 apps also include Read Aloud in Word, Outlook and other tools, helpful for proofreading and accessibility. Edge's Read Aloud is free and highlights words as it reads. Azure AI Speech supports custom voices, speaking styles and SSML markup for controlling pronunciation and pauses. GitHub Copilot, by contrast, is a coding assistant and does not generate speech, though it can write code that calls speech APIs. Check language availability, since neural voice coverage differs across languages.
Is speech to text available in Word?
Yes. Microsoft Word for Microsoft 365 includes Dictate, found on the Home tab, which transcribes your speech into the document. On Windows, voice typing also works in Word and other apps by pressing Windows key plus H.
Dictate needs an internet connection and a Microsoft 365 subscription or account, supports many languages and lets you speak punctuation, such as comma or new line. Word for the web also offers Dictate and a Transcribe feature for recorded audio. For better accuracy, use a good microphone, speak clearly and review the text afterwards. Word's Read Aloud does the reverse, reading text to you, which helps catch errors.
How do you use speech to text in PowerPoint?
In PowerPoint for Microsoft 365, select a text box or the notes pane and click Dictate on the Home tab, then speak. On Windows, you can also press Windows key plus H to use system voice typing in any text field.
Also asked as: how to speech to text in powerpoint
Dictation suits drafting speaker notes and slide text quickly. During presentations, PowerPoint can show live subtitles, transcribing your speech on screen in the same or another language, enabled through the Slide Show tab's subtitle settings. This helps audiences with hearing impairments and multilingual rooms. Test subtitles before presenting, since accuracy depends on microphone quality and room noise. Review dictated text carefully for errors in names and technical terms.
Should speech to text be hyphenated?
Hyphenate it when it works as a compound adjective before a noun, as in speech-to-text software. When used as a phrase describing conversion, as in convert speech to text, leave out the hyphens.
Many technical writers also hyphenate speech-to-text when using it as a noun naming the technology, as in improvements in speech-to-text, and this usage is widely accepted. The same rule applies to text-to-speech. Consistency matters more than any single choice, so follow your style guide and apply it everywhere, including headings. Abbreviations such as STT and TTS are common in technical writing after the first full mention, but spell them out for general audiences.
Sources
GenAI engineer and AI product builder with 2+ years shipping production-grade AI systems: RAG pipelines, fine-tuned LLMs, hybrid retrieval and multi-modal apps across vision, speech and OCR, architected end to end from ingestion to deployment. Leads engineering for SCULT INDIA's 14-member team, founded the 500+ member TechVerse Enclave community and has mentored 200+ students. Three hackathon first prizes: Changethon 2025 (IIT Roorkee), Product Genesis at Vividhotsava 2025 (CSJMU) and BYTEBATTLE (MeetKats).
Open to GenAI roles, hackathon judging, mentorship sessions and guest talks at colleges. For campus sessions the inviting institution covers the visit end to end: travel (a First AC train ticket for journeys up to five hours, a flight for anything longer), accommodation, meals, and pickup and drop; the honorarium and workshop fees are negotiable on WhatsApp or on the call, and requests are welcome from institutions that can cover the visit this way. Email pranjulrathour41@gmail.com.
