H3 Audio Perception Producer:
The Audio Perception Producer provides host-configured speech recognition for short English audio, using CPU processing to generate candidate transcriptions. Speech recognition, also called ASR, converts spoken words into text that can help you understand dialogue or narration when preparing an AI art workflow. The documented configuration uses the whisper_large_v3_cpu_en_v1 profile with the comfyui_native route. Its important distinction is that recognized speech remains uncertain evidence rather than a guaranteed record of what was said, so you should review wording before relying on it for creative decisions. Processing depends on the perception services configured by your ComfyUI host; the supplied code does not establish standalone model installation, support for other languages, or additional capabilities such as sound-effect recognition and speaker identification.
H3 Audio Perception Producer Input Parameters:
media
Required input of type H3_MEDIA_PRODUCER_RESULT. This supplies the media producer result to be processed for audio perception. Connect a compatible upstream media producer rather than a plain text transcript or an unrelated audio socket type. The node’s documented task is short English speech recognition, so the supplied media needs to make suitable audio available to the configured host service. The provided interface does not show the internal media structure, accepted file formats, maximum recording duration, or sample-rate requirements. Minimum, maximum, and default values are not specified; this is a connection-based input rather than a numeric setting.
route
Required COMBO input that selects the processing route. The node description explicitly recommends comfyui_native for its short English CPU speech-recognition configuration. This choice matters because the host must provide the corresponding processing service; choosing a route is not evidence that the service is installed or available. The supplied class does not show the complete option list or default value, so no other route choices or fallback behavior can be confirmed. Numeric minimum and maximum values do not apply.
profile_id
Required STRING input identifying the host-configured perception profile. The documented profile is whisper_large_v3_cpu_en_v1, which is intended for short English speech recognition on the CPU. Use the exact identifier so the host can select the intended configuration; an arbitrary model name is not documented as a substitute for a registered profile. The supplied source does not expose the profile’s internal settings, an allowed-profile list, or the field’s default value. Numeric minimum and maximum values do not apply.
device
Required COMBO input for selecting the processing device. The documented capability is CPU-based speech recognition, so use the CPU selection available in your installation for the recommended profile. The supplied interface does not reveal the literal dropdown entries, the default selection, or whether any other devices are accepted. GPU support or acceleration cannot be inferred from this node description. Numeric minimum and maximum values do not apply.
cancel_requested
Required BOOLEAN input representing a cancellation request. Its possible values are true and false; for an ordinary run where you do not intend to request cancellation, use false. The inherited implementation is not included, so the supplied code does not establish when cancellation is checked, whether it can interrupt active recognition, or what result is returned after cancellation. The default value is not shown, and numeric minimum and maximum values do not apply.
local_service_consent
Optional BOOLEAN input representing consent associated with local-service use. Its possible values are true and false, but the supplied audio-node source does not explain when this consent is checked or whether the recommended comfyui_native configuration requires it. The explicit local-transfer consent requirement in the context belongs to the visual producer and should not automatically be applied to this audio node. Enable consent only when your host documents the relevant service and you agree to its handling of your media. The default and the behavior when this input is omitted are not specified; numeric minimum and maximum values do not apply.
H3 Audio Perception Producer Output Parameters:
result
Output of type H3_AUDIO_PRODUCER_RESULT. This is the audio producer’s structured result, intended for a compatible downstream connection rather than direct use as a plain text string. The node description establishes that any transcript candidates remain uncertain. A supplied host-adapter excerpt also demonstrates an audio-analysis batch marked PARTIAL containing a transcript observation labeled en; this shows that a transcript observation can be available without representing complete audio analysis. However, that excerpt does not define every possible result or expose the complete output structure. Treat recognized words as candidate evidence, review important passages against the recording, and do not assume that this output includes speaker identities, sound-event labels, timestamps, or a complete account of all audio content.
H3 Audio Perception Producer Usage Tips:
- Start with the documented configuration: set
routetocomfyui_native, enterwhisper_large_v3_cpu_en_v1asprofile_id, and select the available CPU device option. - Confirm that your ComfyUI host has the recommended profile and processing service configured before troubleshooting recognition quality. The node explicitly depends on host configuration.
- Use short English speech recordings for initial tests. The supplied source gives no exact duration limit, so check your host’s profile documentation before submitting longer material.
- Connect
mediafrom an output with the exact typeH3_MEDIA_PRODUCER_RESULT, and connectresultonly to a downstream input that acceptsH3_AUDIO_PRODUCER_RESULT. - Keep
cancel_requestedset tofalsefor a normal test run. The supplied source does not document cancellation timing or recovery behavior. - Review candidate transcriptions before incorporating them into prompts, captions, or narrative decisions, especially when a single misheard name or phrase could change your intended meaning.
- Do not assume that changing
deviceenables GPU acceleration or that changing the profile name adds language support. Only short English CPU ASR is documented here. - Check your host’s consent requirements before changing
local_service_consent; do not treat it as a universal switch for fixing configuration problems.
H3 Audio Perception Producer Common Errors and Solutions:
Host perception service unavailable — diagnostic condition, not a verified error string
- Explanation: The node describes its speech recognition as host-configured. The supplied excerpts do not include the inherited execution logic or literal exception messages, so the exact message for an unavailable service cannot be quoted.
- Solution: Check your host’s setup instructions and logs to confirm that the native audio-perception service and recommended profile are available.
Profile not found or unsupported — diagnostic condition, not a verified error string
- Explanation: A profile identifier must correspond to a configuration recognized by your host. The documented identifier is
whisper_large_v3_cpu_en_v1; additional profiles are not listed in the supplied source. - Solution: Check the spelling of
profile_idand ask your host administrator to verify that this exact profile is configured.
Route or device configuration mismatch — diagnostic condition, not a verified error string
- Explanation: A configuration outside the documented
comfyui_nativeroute and CPU processing path is not confirmed to work. The supplied class does not expose the full compatibility rules or dropdown choices. - Solution: Return to
comfyui_native,whisper_large_v3_cpu_en_v1, and the available CPU device selection, then test with a short English recording.
Incompatible socket connection — diagnostic condition, not a verified error string
- Explanation: This node requires a structured
H3_MEDIA_PRODUCER_RESULTconnection and returnsH3_AUDIO_PRODUCER_RESULT. A generic audio connection or plain text connection is not interchangeable with these declared types. - Solution: Use an upstream producer with the required media output type and a downstream node with the matching audio producer input type.
No usable transcript or inaccurate recognized words — result-quality symptom, not a verified error string
- Explanation: Transcript candidates are explicitly uncertain, and the documented scope is short English speech. A missing or inaccurate transcription does not by itself identify whether the cause is the recording, host configuration, or recognition limitations.
- Solution: Confirm that the supplied media contains audible English speech, test with a short clear recording, and compare any recognized text with the original audio before using it.
Partial audio analysis — documented status, not necessarily an error
- Explanation: The host-adapter excerpt includes
AudioAnalysisStatus.PARTIALalongside a transcript observation. A partial status can therefore coexist with available transcript evidence and should not automatically be interpreted as total failure. - Solution: Inspect the available observations through a compatible downstream node and use only the evidence actually provided. Check host logs if you expected additional analysis.
Cancellation requested — diagnostic condition, not a verified error string
- Explanation: The interface exposes
cancel_requested, but the provided source does not specify how cancellation affects execution or the returned result. - Solution: Set
cancel_requestedtofalsewhen you intend to process the recording, then rerun. Consult your host’s logs for the actual cancellation behavior.
Local-service consent required — host-dependent condition, not a verified audio-node error
- Explanation: Although
local_service_consentis present, the supplied audio-node code does not establish a consent requirement for the recommended route. - Solution: If your host explicitly reports a consent requirement, review the service and media-handling details before granting consent. Do not enable it solely on the assumption that every audio failure is consent-related.
