Kovati Docs

The toolset

The whole SpeechForge pipeline, exposed to an agent.

Documented in depthSpeechForgeToolset0.2.2Open source

SpeechForge has no toolset of its own. The sibling plugin exposes the whole pipeline, and adds nothing.

Coverage, grouped by what it is for — and it is the whole panel, not a subset.

GroupTools
DiscoveryListSpeechAssets, ListSpeakers, ListVoiceProfiles, ListSpeechProviders, ListSpeechTranslationProviders, ListProviderVoices
StatusGetSpeechLineStatus, GetSpeechBankSource, CheckSpeechSourceDrift, DetectEditedSpeechAudio, GetSpeechLocalizationStatus
AuthoringCreateOrUpdateSpeechBank, UpdateSpeechLine, RemoveSpeechLines, ClearSpeechBank
CastingCreateOrUpdateSpeaker, CreateVoiceProfile, SetSpeechLineVoiceOverride, ResolveSpeechVoice
CostEstimateSpeechCost, EstimateSpeechConversionCost
GenerationGenerateSpeech, GenerateSpeechTake, ProduceSpeechBank, RefetchSpeechLine
ConversionConvertSpeechLine, ConvertSpeechTake
TakesListSpeechTakes, ApplySpeechTake, MarkSpeechTakeChosen, SetSpeechTakeVariant, RemoveSpeechTake, RemoveSpeechTakeVariant
GraduationMarkSpeechLineRecorded, RegisterRecordedSpeechTake, ApplyRecordedAudio, AcceptCurrentSpeechAudio
LocalisationLocalizeSpeechBank, DubSpeechLine
Providers and pathsGetSpeechProviderCapabilities, GetSpeechCredentialStatus, TestSpeechProviderConnection, GetSpeechOutputPaths, SetSpeechOutputRoot
NavigationOpenSpeechLibrary

ResolveSpeechVoice is the one to reach for when a voice is wrong. It reports which voice a line would actually be spoken in and why — which of the four resolution steps answered. Without it, a wrong voice is an archaeology expedition.

PluginRoleLicenceVersion
SpeechForge sourcecoreThe subsystem. Install this one.Open0.2.2
SpeechForgeDeepLproviderA route to a generator — a vendor API, or a container on your own GPU.Paid0.1.2
SpeechForgeElevenLabs sourceproviderA route to a generator — a vendor API, or a container on your own GPU.Open0.1.1
SpeechForgeToolset sourcetoolsetThe same subsystem as typed MCP tools, for an agent. Adds nothing of its own.Open0.2.2

Casting from an agent

The casting tools are the ones worth knowing about, because a voice cast by an agent and a voice cast by a person must end up identical:

  • Create a voice profile, create or update a speaker, list either.
  • Set a per-line voice override, update a line.
  • Dedupe a remote voice into a profile — ask for a profile for a provider's voice and get the existing one if there is one, rather than a second profile pointing at the same voice.

That last one exists because an agent casting forty speakers from a vendor library will otherwise create forty profiles for the eight voices it actually used.

Migration is a tool, not a redirect

When the casting model changed, the old assets were deleted and the data migrated live through the toolset rather than carried forward on redirects.

That is the general rule here: redirects are migration scaffolding. Bake them in by resaving, then delete them. A project carrying redirects for a rename that happened months ago is carrying a liability nobody remembers.

What an agent may not do

  • Write a credential. It can report whether one is present and where it was read from.
  • Overwrite a recording. A forced regeneration refuses anything whose origin is not Generated, and the refusal says which.
  • Create or delete a voice on the provider. A key that can create one can generally delete one, and deleting a voice loses every line it ever spoke, unrecoverably. There is no tool for it at all.

The second one is the whole point of the two-axis graduation model. An agent asked to "regenerate everything stale" must not be able to destroy a line an actor recorded.

Two estimate tools, because two things bill differently. EstimateSpeechCost prices synthesis per character and can quote a figure. EstimateSpeechConversionCost prices re-voicing, which bills by the second of audio that does not exist yet. The panel's own estimate line says both halves separately for the same reason.

Every version of SpeechForgeToolset.

On this page