Text to speech API: turn text into natural voice audio

Voice Studio is the text-to-speech tool inside the apillms console. You paste text, pick a voice, and get back an audio file without writing any code. This post walks through the full workflow — picking a voice, watching the character counter, understanding what a job costs, downloading from the media library, and a few troubleshooting notes for longer scripts.

What Voice Studio does

Voice Studio is one of the console pages after you sign in, alongside Playground, Image studio, and Video studio. It's built for quick, one-off text-to-speech jobs: paste a script, pick a voice, generate, download. There's no project setup and no API call to write yourself — the console sends the request for you and shows you the result in the browser.

If you'd rather call text-to-speech from code or from an agent instead of a browser tab, the same capability is exposed as a tool on the apillms MCP server (text_to_speech and list_voices). Voice Studio is the fastest path when you just want the audio file right now and don't need it wired into a pipeline.

Step 1: pick a voice

On the Voice Studio page you pick a voice from a list before entering any text. Different voices suit different use cases — narration for a video, a product demo voiceover, a short notification clip for an app, a sample line for a podcast intro. Because voices sound different in tone and pacing, it's worth running a short test line (a sentence or two) through two or three candidates before committing to one for a long script.

A practical pattern: type one short sentence, generate it with voice A, note which one it sounds like, switch to voice B, regenerate the same sentence, and compare. This costs very little in credits (see the cost section below) and saves you from generating an entire script in the wrong voice.

Step 2: paste your text and watch the character counter

Once you've picked a voice, paste or type the text you want spoken into the text box. As you type or paste, Voice Studio shows a running character count. This matters because text-to-speech on apillms is billed per character, not per token like chat models — the counter is the most direct signal of what a job will cost before you hit generate.

A few things worth doing before you generate a long script:

• Paste the full script first and check the counter — long documents, transcripts, or blog posts can run into the tens of thousands of characters quickly.
• If the counter looks higher than expected, scan for leftover formatting (HTML tags, markdown symbols, extra whitespace) that got pasted in along with the text — those characters are billed too, even though they're not spoken.
• Split very long scripts into sections (for example, by chapter or scene) if you want to review and regenerate individual parts without redoing the whole thing.

What text-to-speech costs

Text-to-speech on apillms is priced at 6.37 credits per 1,000 characters. One credit equals $0.005, and credits come from the packs on the Billing page: Starter $5 (1,000 credits), Builder $20 (4,200 credits), Pro $50 (11,000 credits), Scale $100 (23,000 credits).

New accounts get 50 free credits to start, which is enough to try Voice Studio — pick a voice, generate a few test lines, see how it sounds — before buying a pack. Credits never expire, so a pack you buy for voice work can also cover chat, image, or video jobs later, and vice versa.

Use this rate to estimate longer jobs yourself: a 5,000-character script is roughly five times the first row, a 50,000-character script roughly five times the second row. The character counter in Voice Studio gives you the exact length before you commit, so you don't need to guess — count, multiply by the rate above, and you know the cost before generating.

Failed requests are not billed, so a job that errors out doesn't eat into your balance — you only pay for audio that actually generates. For a full cost breakdown of other model types — chat, image, and video — see the pricing page or compare apillms against OpenRouter on /compare.

Step 3: generate and listen

Once you've picked a voice and pasted your text, click generate. The output is an audio file you can play back right in Voice Studio to check it sounds the way you want before you move on.

If something's off — wrong pacing, a mispronounced word, the voice doesn't fit the content — the fix is usually to adjust the text (spell out an abbreviation, break a long sentence into two, add punctuation where you want a pause) and regenerate rather than trying to fix the audio itself, since Voice Studio doesn't have audio-editing controls beyond the voice choice.

Step 4: find your audio in the Media library

After generating, the audio is saved to the Media library console page along with any images or videos you've generated elsewhere in the console. The media library keeps generated files for a limited time, so download anything you want to keep rather than relying on it as long-term storage.

A simple habit that avoids losing work: download the file right after you confirm it sounds right, rename it to something meaningful on your own machine, and treat the media library as a temporary staging area rather than your archive.

Checking usage after you generate

If you want to see how much your voice jobs have cost over time, the Usage console page breaks spend down by product (chat, image, video, voice), by time range (today, 7, or 30 days), and lets you filter by type and status and export everything to CSV. That's the easiest way to confirm the character counter in Voice Studio lined up with what you were actually billed, especially if you ran several test generations while picking a voice.

If you're generating audio from an agent rather than the console, the MCP server's balance tool gives you the same kind of check programmatically — see the MCP server post for how that fits into an automated workflow alongside text_to_speech and list_voices.

Using text-to-speech from an agent instead of the console

Voice Studio is the quickest way to generate a one-off clip by hand, but if you want audio generation as part of a larger pipeline — for example, a Claude Code or Codex workflow that writes a script and then reads it aloud — the MCP server exposes the same functionality as tools. Adding the server in Claude Code looks like:

claude mcp add --transport http apillms https://apillms.com/mcp --header "Authorization: Bearer YOUR_KEY"

Once connected, an agent can call list_voices to see the available options and text_to_speech to generate audio, using the same per-character pricing described above. This is also the approach to use if you're building an automation in n8n or another tool that can call the MCP server or the REST endpoints directly, rather than clicking through the console for every clip.

Troubleshooting

• The character count is higher than I expected. Check for pasted formatting (HTML, markdown, extra line breaks) sitting in the text box — those characters count toward billing even though a listener never hears them as words.
• I generated the wrong voice by mistake. Failed requests aren't billed, but a successful generation in the wrong voice is — so confirm the voice selection before generating a long script, not after.
• My audio file disappeared. The media library keeps files for a limited time only. Download anything you need to keep as soon as you've confirmed it's right.
• I'm not sure what a job will cost before I run it. Watch the character counter as you paste — at 6.37 credits per 1,000 characters, you can work out the cost of any length before clicking generate.

Takeaway

Voice Studio turns text into speech in four steps: pick a voice, paste text and watch the character counter, generate and listen, then download from the media library before it expires. At 6.37 credits per 1,000 characters (1 credit = $0.005), a short test clip costs very little — generate a quick sample first to confirm the voice fits, then run the full script once you're confident in the choice. For automated or recurring voice generation, the same tool is available through the MCP server instead of the console.

FAQ

How is text-to-speech priced on apillms?

It's billed per character at 6.37 credits per 1,000 characters, where 1 credit equals $0.005. The character counter in Voice Studio shows the length before you generate, so you can estimate the cost ahead of time.

Where do my generated audio files go?

They're saved to the Media library console page along with generated images and videos. The library keeps files for a limited time, so download what you want to keep.

Can I use text-to-speech from code instead of the console?

Yes. The apillms MCP server exposes a text_to_speech tool and a list_voices tool, so agents and scripts can generate audio the same way Voice Studio does, without clicking through the browser.

Do failed Voice Studio jobs cost credits?

No, failed requests are not billed on apillms — you only pay for audio that actually generates.