Struct NamModel 

pub struct NamModel {
    pub version: String,
    pub architecture: String,
    pub config: ModelConfig,
    pub weights: Vec<f32>,
    pub sample_rate: Option<f64>,
    pub metadata: Option<Value>,
}

A parsed .nam model file.

This is the file representation — the raw config + flat weight blob. To run inference, build a crate::WaveNet from it.

Fields

version: String

.nam format version string (e.g. "0.5.4").

architecture: String

Model architecture, e.g. "WaveNet".

config: ModelConfig

Architecture-specific configuration (dispatched on Self::architecture).

weights: Vec<f32>

Flat weight blob. The final element is head_scale (see NAM export_weights). Stored as f32 to match NAM Core’s inference precision.

sample_rate: Option<f64>

Training sample rate. Absent in older files; see Self::expected_sample_rate.

metadata: Option<Value>

Opaque training/gear metadata. Not used for inference.

Implementations

impl NamModel

pub fn from_file(path: impl AsRef<Path>) -> Result<Self, Error>

Read and parse a .nam model from a file on disk.

Convenience over std::fs::read_to_string + Self::from_json_str. Returns Error::Io if the file can’t be read, or Error::Json if its contents aren’t valid .nam JSON.

pub fn from_json_str(json: &str) -> Result<Self, Error>

Parse a .nam model from a JSON string already in memory.

pub fn expected_sample_rate(&self) -> f64

The sample rate, in Hz, the model expects its input to be at — falling back to DEFAULT_SAMPLE_RATE when the file does not specify one.

You must feed the model audio at this rate. nam-rs runs the forward pass at whatever rate you hand it and does not resample. A model captured at one rate fed audio at another produces silently wrong output: its dilations and recurrence are defined in samples, not seconds. If your host runs at a different rate, resample to this rate before crate::Model::process_buffer and back afterwards — resampling is the caller’s responsibility. Mirrors NAM Core’s GetExpectedSampleRate().

pub fn metadata_typed(&self) -> Metadata

The file’s descriptive Metadata — name, gear, tone, trainer, date, loudness and calibration levels — parsed from the raw metadata block in one shot.

Returns defaults (every field None) when the file has no metadata block. Fields the block omits are None; keys we don’t type are ignored, and remain reachable on Self::metadata. Parsing is per-field lenient, so a single malformed entry costs only itself.

Prefer this over the single-field accessors (Self::loudness, etc.) when you want several fields: each single-field accessor re-clones and re-parses the raw JSON, whereas this parses once. (All are cold-path / load-time, so neither is on the audio thread.)

pub fn loudness(&self) -> Option<f32>

Output loudness in dBFS RMS, if the file records it — see Metadata::loudness, and note it is not LUFS.

pub fn input_level_dbu(&self) -> Option<f32>

Input calibration level in dBu (analog level at 0 dBFS in), if present.

pub fn includes_cab(&self) -> Option<bool>

Whether this capture already includes a speaker cabinet — see Metadata::includes_cab for the details and the None case.

Use this to decide whether to put an impulse response after the model: a cab-inclusive capture followed by an IR puts two cabs in series.

pub fn output_level_dbu(&self) -> Option<f32>

Output calibration level in dBu (analog level at 0 dBFS out), if present.

Trait Implementations

impl Clone for NamModel

fn clone(&self) -> NamModel

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for NamModel

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for NamModel

fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>

where D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.