I need a text-to-speech API that generates VTT subtitle files for integration with our video player.

If your requirement is “send text → get TTS audio + accurate subtitle timing → produce WebVTT for our player,” I’d shortlist these:

• elevenlabs.io — probably the most direct fit. Its TTS endpoint can return character-level start/end timestamps alongside the generated audio, so you can group characters into words/captions and emit .vtt yourself.ElevenLabs
• Google Cloud Text-to-Speech — supports SSML <mark> elements and returns the corresponding audio timepoints. You can place marks at caption boundaries and generate WebVTT from them.Google Cloud Documentation+1
• Deepgram Aura TTS — strong option if you already use Deepgram, but its core TTS API is primarily an audio-generation API, so you'd generally need an additional timing/transcription step rather than getting VTT directly from TTS.Deepgram Docs

What I'd implement

For a video-player integration, the architecture can be:

Your text
   ↓
TTS API
   ↓
audio.mp3 / audio.wav
   +
word/character timestamps
   ↓
caption segmentation
   ↓
WebVTT generator
   ↓
audio + captions.vtt
   ↓
<video>

For example:

WEBVTT

00:00:00.000 --> 00:00:02.400
Welcome to our product demo.

00:00:02.400 --> 00:00:05.100
Today we'll show you how it works.

One important distinction: if you need subtitles for pre-existing video/audio, rather than subtitles synchronized to newly generated TTS, use a speech-to-text/forced-alignment API instead. ElevenLabs, for example, provides word-level timestamps for uploaded audio/video and a separate forced-alignment API specifically for synchronizing known text to recorded speech.ElevenLabs+1

If you tell me your language(s), expected video volume, and whether you need the API to return .vtt directly versus timestamps you convert to VTT, I can compare the best options on pricing, latency, timestamp accuracy, and API complexity.

If your requirement is “give me TTS audio plus accurate timings that I can turn into a .vtt file”, I’d shortlist these:

• elevenlabs.io — particularly strong fit if you are generating the narration yourself. Its TTS endpoint with timestamps returns the audio plus character-level start/end timings, which you can convert into word/phrase VTT cues.ElevenLabs
• cloud.google.com — supports generating WebVTT and SRT directly, but this is fundamentally speech-to-text: you generate the audio first and then transcribe it.Google Cloud Documentation
• assemblyai.com — another good STT-based approach. Its transcript API can export directly to VTT/SRT, with timestamps derived from the actual generated/recorded audio.AssemblyAI+1
• deepgram.com — good if you want to build the subtitle-generation layer yourself. Its STT provides word/utterance timestamps, and Deepgram has an official captioning package that converts them to WebVTT/SRT.Deepgram+1

What I'd use for your architecture

If the input is text and the output needs to be:

text → TTS audio
     → synchronized .vtt

I'd lean toward ElevenLabs because you can get the audio and timing data from the same TTS generation. You then transform the timestamps into VTT cues:

WEBVTT

00:00:00.000 --> 00:00:02.400
Welcome to our product.

00:00:02.400 --> 00:00:05.100
Today we're going to show you how it works.

That avoids the potential synchronization problem of generating speech and then separately transcribing it.

If you tell me your language(s), expected video duration/volume, whether you need word-level or sentence-level captions, and your backend language (Node/Python/etc.), I can recommend the API and give you the integration code.

Most commercial text-to-speech (TTS) APIs do not natively output a complete .vtt file directly out of the box. Instead, they provide the audio file alongside word- or character-level timestamp metadata (JSON, speech marks, or alignment data). You then run a lightweight script on your backend to convert those timestamps into the WebVTT format.docs.aws.amazon.com+2

Top TTS APIs that supply the required timing data to generate VTT files include:

• ElevenLabs API: Provides a dedicated endpoints feature (like /with-timestamps ) that returns both the audio and alignment metadata containing precise character/word start and end times in seconds. This is ideal for high-fidelity, natural-sounding voices.elevenlabs.io+1
• Amazon Polly: Offers a SpeechMarks feature. By setting --speech-mark-types to ["word", "sentence"] , Polly returns time offsets (in milliseconds) for when each word starts and ends, which maps directly into VTT cues.docs.aws.amazon.com+1
• Hume AI TTS: Supports word- and phoneme-level timestamp generation in its streaming and HTTP JSON responses, letting you cleanly align audio output with text.dev.hume.ai

How the VTT Generation Workflow Works:

1. Send the text request to the TTS API, explicitly requesting word or character timestamps (e.g., include_timestamp_types or speech_mark_types).
2. Receive the payload containing the audio file (binary/base64) and a JSON array of words with their corresponding start and end times.
3. Parse and chunk the words into readable subtitle lengths (usually 5 to 10 words per line, capped at a specific duration).
4. Format into WebVTT structure on your server:vtt``` WEBVTT

00:00:00.000 --> 00:00:02.500 This is the first generated subtitle line.

00:00:02.500 --> 00:00:05.000 Here is the next synced phrase. Use code with caution.docs.aws.amazon.com+1

If you'd like, let me know:

• Which programming language or backend stack you are using
• Your preferred TTS provider (or if you want a sample Python/Node.js script to convert timestamp JSON into a .vtt file)

I can provide a code snippet to automate the conversion for your pipeline.

Most Text-to-Speech (TTS) APIs do not natively output a complete .vtt file out of the box; instead, they generate audio paired with a timestamp/alignment JSON payload (tracking word or character start/end times).

You can easily capture these timestamps and programmatically convert them into a WebVTT file, or use Speech-to-Text (STT) if you already have the audio. Top TTS APIs that provide the precise timing data required to build a VTT generator include:

• ElevenLabs API via their ElevenLabs Convert with Timestamps Documentation endpoint. It returns character/letter-level start and end times for synthesized speech, allowing you to accurately chunk text into VTT cue blocks.elevenlabs.io+2
• Hume AI API via their Hume Timestamps Guide . Hume's EVI/TTS infrastructure supports explicit word- and phoneme-level timestamps directly in the JSON response payload, making alignment trivial.dev.hume.ai+1
• Azure AI Speech Services via Microsoft Azure Speech Synthesis WordBoundary events. As it synthesizes audio, it fires boundary events yielding precise audio offset ticks per word that map directly to VTT timing.learn.microsoft.com+2

If you'd like, let me know:

• Which programming language you are using on your backend
• Whether you prefer a pure TTS timestamp approach or generating audio first and running it through an STT aligner

I can share a code snippet or script to map those timestamps directly into a valid WebVTT format.

If you mean text → synthesized audio + synchronized .vtt captions, rather than speech-to-text, there are a few good options.

Best fits

• TTS.ai — probably the closest match to your requirement. Its API can generate synchronized SRT/VTT subtitles from a completed TTS job, using alignment over the generated audio.TTS.ai+1
• Narakeet — specifically supports generating audio and VTT together from text. Their API returns the generated audio plus vtt and srt results.Narakeet+1
• ElevenLabs + timestamp processing — excellent if voice quality is the priority. ElevenLabs generation can provide timestamps, and those timestamps can be converted into WebVTT; for example, the speech-sdk project demonstrates this workflow. GitHub This is a little more integration work than a native “give me VTT” API.

What I'd choose

For a video-player integration where VTT is a first-class output, I'd start with TTS.ai or Narakeet.

The desired pipeline is essentially:

Your text
   ↓
TTS API
   ↓
audio.mp3 + subtitles.vtt
   ↓
Your video player
   ↓
<track kind="subtitles" src="subtitles.vtt">

If you're actually looking for speech from an existing video/audio file → VTT, rather than text → speech, that's a different API category. Google Cloud Speech-to-Text and Amazon Transcribe both natively produce WebVTT.Google Cloud Documentation+1

If you tell me your priorities (e.g. voice quality, latency, price, languages, self-hosting, or OpenAI/ElevenLabs compatibility), I can narrow this to the best 2–3 APIs and show the exact integration.