Respeecher connects performed speech, licensed voices and streaming TTS

Respeecher supplies synthetic speech through two distinct production paths: converting a recorded performance into another voice, and generating speech directly from text. Its Marketplace supports creative production around recordings and projects, while Space is a separate streaming text-to-speech API. The right entry point depends on whether a human performance or a written script should determine the result.

01 / ProductPerformance conversion and text generation are different tools

The Marketplace accepts recorded or uploaded speech as well as typed text. Speech-to-speech matters because the source performance already contains timing, emphasis and emotion. A director can guide a performer and then change the vocal identity. Text-to-speech instead asks the model to generate a performance from written instructions and its available controls.

That distinction continues into the API selection guide. Space is the default route for streaming TTS, while Marketplace provides speech conversion and project-based operations, including recordings and calibration. They use separate base URLs, authentication headers and keys. A team should choose the workflow before building an integration around an example request.

Respeecher also offers managed voice work and customized arrangements. A catalog voice, a commissioned voice and a streaming voice are not automatically the same asset with the same rights. The company's ethics material makes permission and responsible use central to its positioning. The reader still needs terms covering the actual performer, purpose and distribution of the finished production.

02 / AudienceA fit for directed media and application builders

A film, game or animation team may value the ability to keep an actor's performance while changing its voice characteristics. In that setting, the meaningful unit of quality is a line delivered in context: its relationship to the preceding scene, the timing of the response and the intention behind an emphasized word. A convincing isolated sample is useful discovery evidence, but not a finished casting decision.

An application developer has another requirement. The service needs speech as text becomes available, must recover from interruptions and has to maintain the same speaker identity across requests. Space explicitly targets that streaming use case. Its published latency language is a vendor claim, so responsiveness should be measured through the complete application and the network users will actually use.

Respeecher is less directly suited to a team whose main problem is editing long recordings and managing the surrounding video. Descript addresses an editing-centered workflow. ElevenLabs provides another speech-generation and audio-production approach. Compare these against the same scene or script rather than treating all tools with a voice library as equivalent.

03 / WorkflowA proposed character-dialogue production

Consider a proposed game-dialogue pilot containing a short exchange between two characters. Record a performer delivering several versions of each line, with permission covering the intended conversion and use. Keep clean source takes and identify the preferred performance before selecting a target voice. This makes it possible to tell whether a problem comes from the original delivery or the conversion.

Use Marketplace's project and recording workflow for this path. The API guide describes calibration for matching source and target pitch, along with conversion orders that are submitted, polled and downloaded. Preserve the chosen recording, voice and settings for each approved line. Replacing a line should be a deliberate new revision, not an unexplained overwrite in an export folder.

Review converted dialogue inside the actual scene, including pauses, background audio and neighboring lines. Ask a director to assess intention and a separate listener to flag words that are hard to understand. Include a quiet sentence, a fast interruption and a proper name. These are proposed test cases, not reported Respeecher defects or evidence of completed hands-on testing.

For a live character that speaks dynamically, run a separate Space experiment instead of assuming the batch production setup will stream. Test cancellation, sentence boundaries and the point at which audio becomes audible. Keep scripted safety-critical instructions prerecorded or otherwise controlled until the team has established how generated phrasing and delivery will be approved.

04 / PricingKeep the two pricing systems apart

Marketplace pricing includes prepaid credits and subscriptions. The displayed starter pack is $5 for five credits, described as either 20,000 TTS characters or five speech-to-speech minutes. The wording is an alternative allowance, not both quantities together. Marketplace terms identify US-dollar payment and prohibit commercial projects on trial plans. They also restrict commercial customers whose annual revenue exceeds US$1 million to a custom subscription; larger studios should not assume a self-service pack is an eligible production route.

The Space page separately lists $2 per 60,000 characters, described as roughly one hour of speech. Budget by characters; speaking speed and punctuation make the hour estimate unsuitable as a guaranteed duration. Space offers top-ups or automatic recharge and custom volume discussions. Its rate should not be applied to Marketplace speech conversion.

For Marketplace subscriptions, the public page displays introductory discounts beside regular monthly and annual figures. This article avoids collapsing those into a single apparent price. Check the selected term and allowance in the account before committing. Also account for discarded generations: the useful financial comparison is the production effort and usage required to obtain an approved scene, not merely the shortest possible sample.

05 / DistinctionsThe source performance can remain the creative anchor

Speech-to-speech creates a different creative control point from text prompts alone. A director can resolve a hesitant delivery or comic pause with the performer first, then evaluate how the converted voice carries it. This does not remove the need for sound editing, but it can make the intended performance explicit before synthesis. The resulting production records should keep both the input take and the approved output.

The split between Space and Marketplace is also more than packaging. The documentation makes an explicit technical choice between immediate TTS streaming and ordered project work. That helps developers avoid adding polling and project management to a task that only needs speech output, while retaining those structures when a production requires them.

Licensed voice sourcing is another consequential distinction. The Space product page describes compensated real performers and revenue sharing. That is a vendor statement about its sourcing model, not proof that every possible use of a voice is allowed. Brand endorsements, character likeness and specific talent restrictions need to be resolved for the production rather than inferred from catalog availability.

06 / QuestionsClarify rights, revision control and continuity

Can the selected voice be used for this specific release? The Marketplace agreement separates trial restrictions, customer inputs and permitted use. Read any additional voice-specific terms before attaching a recognizable identity to a commercial character. The fact that a voice can be auditioned does not grant permission to imply the real person endorsed the project.

How will the production remain reproducible? Save source takes and approved exports in the team's own managed archive, along with voice identifiers and conversion settings. Ask how access to a particular voice is maintained over a long project and what happens if that catalog entry changes. A sequel or localization update can require continuity long after the original production sprint.

What does the API failure path sound like? Silence, an unfinished word and an unexpected speaker change affect users differently. For streaming, define a fallback and measure interruption behavior. For batch work, keep failed conversions visible and avoid treating the existence of a download URL as an editorial approval. The source material reviewed here establishes available workflows, not a measured error rate for either service.

07 / DecisionChoose around the performance you need to preserve

Respeecher deserves a place in a shortlist when directed speech and voice identity are separate creative requirements, or when a developer wants a dedicated streaming TTS route. The first pilot should answer one of those questions clearly. Mixing a character-production test with an interactive-agent benchmark makes it harder to see which part of the offering is actually helping.