diff --git a/doc/Makefile b/doc/Makefile
index 47c0fdcd..05819848 100644
--- a/doc/Makefile
+++ b/doc/Makefile
@@ -29,7 +29,7 @@ pip3:
 	$(PIP_INSTALL) -r ../taskcluster/docs-requirements.txt
 
 npm:
-	npm install jsdoc@3.6.3
+	npm install typedoc@0.17.4 typescript@3.8.3 @types/node@13.9.x
 
 submodule:
 	git submodule update --init --remote
diff --git a/doc/NodeJS-API.rst b/doc/NodeJS-API.rst
index 7593150c..b6170b5b 100644
--- a/doc/NodeJS-API.rst
+++ b/doc/NodeJS-API.rst
@@ -1,8 +1,6 @@
 JavaScript (NodeJS / ElectronJS)
 ================================
 
-Support for TypeScript is :download:`provided in index.d.ts<../native_client/javascript/index.d.ts>`
-
 Model
 -----
 
diff --git a/doc/conf.py b/doc/conf.py
index 27dc16e1..3557c586 100644
--- a/doc/conf.py
+++ b/doc/conf.py
@@ -80,7 +80,9 @@ breathe_projects = {
   "deepspeech-dotnet": "xml-dotnet/",
 }
 
-js_source_path = "../native_client/javascript"
+js_source_path = "../native_client/javascript/index.ts"
+js_language = "typescript"
+jsdoc_config_path = "../native_client/javascript/tsconfig.json"
 
 # Add any paths that contain templates here, relative to this directory.
 templates_path = ['.templates']
diff --git a/native_client/javascript/Makefile b/native_client/javascript/Makefile
index b634cf4b..28b00046 100644
--- a/native_client/javascript/Makefile
+++ b/native_client/javascript/Makefile
@@ -59,7 +59,7 @@ copy-deps: build
 node-wrapper: copy-deps build
 	PATH="$(NODE_MODULES_BIN):${PATH}" $(NODE_BUILD_TOOL) $(NODE_PLATFORM_TARGET) $(NODE_RUNTIME) $(NODE_ABI_TARGET) $(NODE_DIST_URL) --no-color package $(NODE_BUILD_VERBOSE)
 
-npm-pack: clean package.json index.js npm-dev
+npm-pack: clean package.json index.ts npm-dev
 	PATH="$(NODE_MODULES_BIN):${PATH}" tsc && $(NPM_TOOL) pack $(NODE_BUILD_VERBOSE)
 
 deepspeech_wrap.cxx: deepspeech.i ds-swig
diff --git a/native_client/javascript/client.ts b/native_client/javascript/client.ts
index 7d762d66..53c9eac4 100644
--- a/native_client/javascript/client.ts
+++ b/native_client/javascript/client.ts
@@ -3,7 +3,7 @@
 // This is required for process.versions.electron below
 /// <reference types="electron" />
 
-import Ds from "./index";
+import * as Ds from "./index";
 import * as Fs from "fs";
 import Sox from "sox-stream";
 import * as argparse from "argparse";
diff --git a/native_client/javascript/index.d.ts b/native_client/javascript/index.d.ts
deleted file mode 100644
index a5d9c514..00000000
--- a/native_client/javascript/index.d.ts
+++ /dev/null
@@ -1,196 +0,0 @@
-/**
- * Stores text of an individual token, along with its timing information
- */
-export interface TokenMetadata {
-    text: string;
-    timestep: number;
-    start_time: number;
-}
-
-/**
- * A single transcript computed by the model, including a confidence value and
- * the metadata for its constituent tokens.
- */
-export interface CandidateTranscript {
-    tokens: TokenMetadata[];
-    confidence: number;
-}
-
-/**
- * An array of CandidateTranscript objects computed by the model.
- */
-export interface Metadata {
-    transcripts: CandidateTranscript[];
-}
-
-/**
- * An object providing an interface to a trained DeepSpeech model.
- *
- * @param aModelPath The path to the frozen model graph.
- *
- * @throws on error
- */
-export class Model {
-constructor(aModelPath: string)
-
-/**
- * Get beam width value used by the model. If :js:func:Model.setBeamWidth was
- * not called before, will return the default value loaded from the model file.
- * 
- * @return Beam width value used by the model.
- */
-beamWidth(): number;
-
-/**
- * Set beam width value used by the model.
- * 
- * @param The beam width used by the model. A larger beam width value generates better results at the cost of decoding time.
- *
- * @return Zero on success, non-zero on failure.
- */
-setBeamWidth(aBeamWidth: number): number;
-
-/**
- * Return the sample rate expected by the model.
- *
- * @return Sample rate.
- */
-sampleRate(): number;
-
-/**
- * Enable decoding using an external scorer.
- *
- * @param aScorerPath The path to the external scorer file.
- *
- * @return Zero on success, non-zero on failure (invalid arguments).
- */
-enableExternalScorer(aScorerPath: string): number;
-
-/**
- * Disable decoding using an external scorer.
- *
- * @return Zero on success, non-zero on failure (invalid arguments).
- */
-disableExternalScorer(): number;
-
-/**
- * Set hyperparameters alpha and beta of the external scorer.
- *
- * @param aLMAlpha The alpha hyperparameter of the CTC decoder. Language Model weight.
- * @param aLMBeta The beta hyperparameter of the CTC decoder. Word insertion weight.
- *
- * @return Zero on success, non-zero on failure (invalid arguments).
- */
-setScorerAlphaBeta(aLMAlpha: number, aLMBeta: number): number;
-
-/**
- * Use the DeepSpeech model to perform Speech-To-Text.
- *
- * @param aBuffer A 16-bit, mono raw audio signal at the appropriate sample rate (matching what the model was trained on).
- *
- * @return The STT result. Returns undefined on error.
- */
-stt(aBuffer: Buffer): string;
-
-/**
- * Use the DeepSpeech model to perform Speech-To-Text and output metadata
- * about the results.
- *
- * @param aBuffer A 16-bit, mono raw audio signal at the appropriate sample rate (matching what the model was trained on).
- * @param aNumResults Maximum number of candidate transcripts to return. Returned list might be smaller than this.
- * Default value is 1 if not specified.
- *
- * @return :js:func:`Metadata` object containing multiple candidate transcripts. Each transcript has per-token metadata including timing information.
- * The user is responsible for freeing Metadata by calling :js:func:`FreeMetadata`. Returns undefined on error.
- */
-sttWithMetadata(aBuffer: Buffer, aNumResults?: number): Metadata;
-
-/**
- * Create a new streaming inference state. One can then call :js:func:`Stream.feedAudioContent` and :js:func:`Stream.finishStream` on the returned stream object.
- *
- * @return a :js:func:`Stream` object that represents the streaming state.
- *
- * @throws on error
- */
-createStream(): Stream;
-}
-
-/**
- * @class
- * Provides an interface to a DeepSpeech stream. The constructor cannot be called
- * directly, use :js:func:`Model.createStream`.
- */
-declare class Stream {
-/**
- * Feed audio samples to an ongoing streaming inference.
- *
- * @param aBuffer An array of 16-bit, mono raw audio samples at the
- *                 appropriate sample rate (matching what the model was trained on).
- */
-feedAudioContent(aBuffer: Buffer): void;
-
-/**
- * Compute the intermediate decoding of an ongoing streaming inference.
- *
- * @return The STT intermediate result.
- */
-intermediateDecode(): string;
-
-/**
- * Compute the intermediate decoding of an ongoing streaming inference, return results including metadata.
- *
- * @param aNumResults Maximum number of candidate transcripts to return. Returned list might be smaller than this. Default value is 1 if not specified.
- *
- * @return :js:func:`Metadata` object containing multiple candidate transcripts. Each transcript has per-token metadata including timing information. The user is responsible for freeing Metadata by calling :js:func:`FreeMetadata`. Returns undefined on error.
- */
-intermediateDecodeWithMetadata (aNumResults?: number): Metadata;
-
-/**
- * Compute the final decoding of an ongoing streaming inference and return the result. Signals the end of an ongoing streaming inference.
- *
- * @return The STT result.
- *
- * This method will free the stream, it must not be used after this method is called.
- */
-finishStream(): string;
-
-/**
- * Compute the final decoding of an ongoing streaming inference and return the results including metadata. Signals the end of an ongoing streaming inference.
- *
- * @param aNumResults Maximum number of candidate transcripts to return. Returned list might be smaller than this. Default value is 1 if not specified.
- *
- * @return Outputs a :js:func:`Metadata` struct of individual letters along with their timing information. The user is responsible for freeing Metadata by calling :js:func:`FreeMetadata`.
- *
- * This method will free the stream, it must not be used after this method is called.
- */
-finishStreamWithMetadata(aNumResults?: number): Metadata;
-}
-
-/**
- * Frees associated resources and destroys model object.
- *
- * @param model A model pointer returned by :js:func:`Model`
- *
- */
-export function FreeModel(model: Model): void;
-
-/**
- * Free memory allocated for metadata information.
- *
- * @param metadata Object containing metadata as returned by :js:func:`Model.sttWithMetadata` or :js:func:`Stream.finishStreamWithMetadata`
- */
-export function FreeMetadata(metadata: Metadata): void;
-
-/**
- * Destroy a streaming state without decoding the computed logits. This
- * can be used if you no longer need the result of an ongoing streaming
- * inference and don't want to perform a costly decode operation.
- *
- * @param stream A streaming state pointer returned by :js:func:`Model.createStream`.
- */
-export function FreeStream(stream: Stream): void;
-
-/**
- * Print version of this library and of the linked TensorFlow library on standard output.
- */
-export function Version(): void;
diff --git a/native_client/javascript/index.js b/native_client/javascript/index.js
deleted file mode 100644
index 4ce77d47..00000000
--- a/native_client/javascript/index.js
+++ /dev/null
@@ -1,336 +0,0 @@
-'use strict';
-
-const binary = require('node-pre-gyp');
-const path = require('path')
-// 'lib', 'binding', 'v0.1.1', ['node', 'v' + process.versions.modules, process.platform, process.arch].join('-'), 'deepspeech-bindings.node')
-const binding_path = binary.find(path.resolve(path.join(__dirname, 'package.json')));
-
-// On Windows, we can't rely on RPATH being set to $ORIGIN/../ or on
-// @loader_path/../ but we can change the PATH to include the proper directory
-// for the dynamic linker
-if (process.platform === 'win32') {
-  const dslib_path = path.resolve(path.join(binding_path, '../..'));
-  var oldPath = process.env.PATH;
-  process.env['PATH'] = `${dslib_path};${process.env.PATH}`;
-}
-
-const binding = require(binding_path);
-
-if (process.platform === 'win32') {
-  process.env['PATH'] = oldPath;
-}
-
-/**
- * @class
- * An object providing an interface to a trained DeepSpeech model.
- *
- * @param {string} aModelPath The path to the frozen model graph.
- *
- * @throws on error
- */
-function Model(aModelPath) {
-    this._impl = null;
-
-    const rets = binding.CreateModel(aModelPath);
-    const status = rets[0];
-    const impl = rets[1];
-    if (status !== 0) {
-        throw "CreateModel failed with '"+binding.ErrorCodeToErrorMessage(status)+"' (0x" + status.toString(16) + ")";
-    }
-
-    this._impl = impl;
-}
-
-/**
- * Get beam width value used by the model. If :js:func:Model.setBeamWidth was
- * not called before, will return the default value loaded from the model file.
- *
- * @return {number} Beam width value used by the model.
- */
-Model.prototype.beamWidth = function() {
-    return binding.GetModelBeamWidth(this._impl);
-}
-
-/**
- * Set beam width value used by the model.
- *
- * @param {number} The beam width used by the model. A larger beam width value generates better results at the cost of decoding time.
- *
- * @return {number} Zero on success, non-zero on failure.
- */
-Model.prototype.setBeamWidth = function(aBeamWidth) {
-    return binding.SetModelBeamWidth(this._impl, aBeamWidth);
-}
-
-/**
- * Return the sample rate expected by the model.
- *
- * @return {number} Sample rate.
- */
-Model.prototype.sampleRate = function() {
-    return binding.GetModelSampleRate(this._impl);
-}
-
-/**
- * Enable decoding using an external scorer.
- *
- * @param {string} aScorerPath The path to the external scorer file.
- *
- * @throws on error
- */
-Model.prototype.enableExternalScorer = function(aScorerPath) {
-    const status = binding.EnableExternalScorer(this._impl, aScorerPath);
-    if (status !== 0) {
-        throw "EnableExternalScorer failed with '"+binding.ErrorCodeToErrorMessage(status)+"' (0x" + status.toString(16) + ")";
-    }
-}
-
-/**
- * Disable decoding using an external scorer.
- *
- * @return {number} Zero on success, non-zero on failure (invalid arguments).
- */
-Model.prototype.disableExternalScorer = function() {
-    return binding.EnableExternalScorer(this._impl);
-}
-
-/**
- * Set hyperparameters alpha and beta of the external scorer.
- *
- * @param {float} aLMAlpha The alpha hyperparameter of the CTC decoder. Language Model weight.
- * @param {float} aLMBeta The beta hyperparameter of the CTC decoder. Word insertion weight.
- *
- * @return {number} Zero on success, non-zero on failure (invalid arguments).
- */
-Model.prototype.setScorerAlphaBeta = function(aLMAlpha, aLMBeta) {
-    return binding.SetScorerAlphaBeta(this._impl, aLMAlpha, aLMBeta);
-}
-
-/**
- * Use the DeepSpeech model to perform Speech-To-Text.
- *
- * @param {object} aBuffer A 16-bit, mono raw audio signal at the appropriate sample rate (matching what the model was trained on).
- *
- * @return {string} The STT result. Returns undefined on error.
- */
-Model.prototype.stt = function(aBuffer) {
-    return binding.SpeechToText(this._impl, aBuffer);
-}
-
-/**
- * Use the DeepSpeech model to perform Speech-To-Text and output results including metadata.
- *
- * @param {object} aBuffer A 16-bit, mono raw audio signal at the appropriate sample rate (matching what the model was trained on).
- * @param {number} aNumResults Maximum number of candidate transcripts to return. Returned list might be smaller than this. Default value is 1 if not specified.
- *
- * @return {object} :js:func:`Metadata` object containing multiple candidate transcripts. Each transcript has per-token metadata including timing information. The user is responsible for freeing Metadata by calling :js:func:`FreeMetadata`. Returns undefined on error.
- */
-Model.prototype.sttWithMetadata = function(aBuffer, aNumResults) {
-    aNumResults = aNumResults || 1;
-    return binding.SpeechToTextWithMetadata(this._impl, aBuffer, aNumResults);
-}
-
-/**
- * Create a new streaming inference state. One can then call :js:func:`Stream.feedAudioContent` and :js:func:`Stream.finishStream` on the returned stream object.
- *
- * @return {object} a :js:func:`Stream` object that represents the streaming state.
- *
- * @throws on error
- */
-Model.prototype.createStream = function() {
-    const rets = binding.CreateStream(this._impl);
-    const status = rets[0];
-    const ctx = rets[1];
-    if (status !== 0) {
-        throw "CreateStream failed with '"+binding.ErrorCodeToErrorMessage(status)+"' (0x" + status.toString(16) + ")";
-    }
-    return new Stream(ctx);
-}
-
-/**
- * @class
- * Provides an interface to a DeepSpeech stream. The constructor cannot be called
- * directly, use :js:func:`Model.createStream`.
- */
-function Stream(nativeStream) {
-    this._impl = nativeStream;
-}
-
-/**
- * Feed audio samples to an ongoing streaming inference.
- *
- * @param {buffer} aBuffer An array of 16-bit, mono raw audio samples at the
- *                 appropriate sample rate (matching what the model was trained on).
- */
-Stream.prototype.feedAudioContent = function(aBuffer) {
-    binding.FeedAudioContent(this._impl, aBuffer);
-}
-
-/**
- * Compute the intermediate decoding of an ongoing streaming inference.
- *
- * @return {string} The STT intermediate result.
- */
-Stream.prototype.intermediateDecode = function() {
-    return binding.IntermediateDecode(this._impl);
-}
-
-/**
- * Compute the intermediate decoding of an ongoing streaming inference, return results including metadata.
- *
- * @param {number} aNumResults Maximum number of candidate transcripts to return. Returned list might be smaller than this. Default value is 1 if not specified.
- *
- * @return {object} :js:func:`Metadata` object containing multiple candidate transcripts. Each transcript has per-token metadata including timing information. The user is responsible for freeing Metadata by calling :js:func:`FreeMetadata`. Returns undefined on error.
- */
-Stream.prototype.intermediateDecodeWithMetadata = function(aNumResults) {
-    aNumResults = aNumResults || 1;
-    return binding.IntermediateDecodeWithMetadata(this._impl, aNumResults);
-}
-
-/**
- * Compute the final decoding of an ongoing streaming inference and return the result. Signals the end of an ongoing streaming inference.
- *
- * @return {string} The STT result.
- *
- * This method will free the stream, it must not be used after this method is called.
- */
-Stream.prototype.finishStream = function() {
-    let result = binding.FinishStream(this._impl);
-    this._impl = null;
-    return result;
-}
-
-/**
- * Compute the final decoding of an ongoing streaming inference and return the results including metadata. Signals the end of an ongoing streaming inference.
- *
- * @param {number} aNumResults Maximum number of candidate transcripts to return. Returned list might be smaller than this. Default value is 1 if not specified.
- *
- * @return {object} Outputs a :js:func:`Metadata` struct of individual letters along with their timing information. The user is responsible for freeing Metadata by calling :js:func:`FreeMetadata`.
- *
- * This method will free the stream, it must not be used after this method is called.
- */
-Stream.prototype.finishStreamWithMetadata = function(aNumResults) {
-    aNumResults = aNumResults || 1;
-    let result = binding.FinishStreamWithMetadata(this._impl, aNumResults);
-    this._impl = null;
-    return result;
-}
-
-
-/**
- * Frees associated resources and destroys model object.
- *
- * @param {object} model A model pointer returned by :js:func:`Model`
- *
- */
-function FreeModel(model) {
-    return binding.FreeModel(model._impl);
-}
-
-/**
- * Free memory allocated for metadata information.
- *
- * @param {object} metadata Object containing metadata as returned by :js:func:`Model.sttWithMetadata` or :js:func:`Stream.finishStreamWithMetadata`
- */
-function FreeMetadata(metadata) {
-    return binding.FreeMetadata(metadata);
-}
-
-/**
- * Destroy a streaming state without decoding the computed logits. This
- * can be used if you no longer need the result of an ongoing streaming
- * inference and don't want to perform a costly decode operation.
- *
- * @param {Object} stream A stream object returned by :js:func:`Model.createStream`.
- */
-function FreeStream(stream) {
-    return binding.FreeStream(stream._impl);
-}
-
-/**
- * Print version of this library and of the linked TensorFlow library on standard output.
- */
-function Version() {
-    return binding.Version();
-}
-
-
-//// Metadata, CandidateTranscript and TokenMetadata are here only for documentation purposes
-
-/**
- * @class
- * 
- * Stores text of an individual token, along with its timing information
- */
-function TokenMetadata() {}
-
-/** 
- * The text corresponding to this token
- *
- * @return {string} The text generated
- */
-TokenMetadata.prototype.text = function() {}
-
-/**
- * Position of the token in units of 20ms
- *
- * @return {int} The position of the token
- */
-TokenMetadata.prototype.timestep = function() {};
-
-/**
- * Position of the token in seconds
- *
- * @return {float} The position of the token
- */
-TokenMetadata.prototype.start_time = function() {};
-
-/**
- * @class
- *
- * A single transcript computed by the model, including a confidence value and
- * the metadata for its constituent tokens.
- */
-function CandidateTranscript () {}
-
-/**
- * Array of tokens
- *
- * @return {array} Array of :js:func:`TokenMetadata`
- */
-CandidateTranscript.prototype.tokens = function() {}
-
-/**
- * Approximated confidence value for this transcription. This is roughly the
- * sum of the acoustic model logit values for each timestep/token that
- * contributed to the creation of this transcription.
- *
- * @return {float} Confidence value
- */
-CandidateTranscript.prototype.confidence = function() {}
-
-/**
- * @class
- *
- * An array of CandidateTranscript objects computed by the model.
- */
-function Metadata () {}
-
-/**
- * Array of transcripts
- *
- * @return {array} Array of :js:func:`CandidateTranscript` objects
- */
-Metadata.prototype.transcripts = function() {}
-
-
-module.exports = {
-    Model: Model,
-    Metadata: Metadata,
-    CandidateTranscript: CandidateTranscript,
-    TokenMetadata: TokenMetadata,
-    Version: Version,
-    FreeModel: FreeModel,
-    FreeStream: FreeStream,
-    FreeMetadata: FreeMetadata
-};
diff --git a/native_client/javascript/index.ts b/native_client/javascript/index.ts
new file mode 100644
index 00000000..d0eb7d81
--- /dev/null
+++ b/native_client/javascript/index.ts
@@ -0,0 +1,307 @@
+import binary from 'node-pre-gyp';
+import path from 'path';
+
+// 'lib', 'binding', 'v0.1.1', ['node', 'v' + process.versions.modules, process.platform, process.arch].join('-'), 'deepspeech-bindings.node')
+const binding_path = binary.find(path.resolve(path.join(__dirname, 'package.json')));
+
+// On Windows, we can't rely on RPATH being set to $ORIGIN/../ or on
+// @loader_path/../ but we can change the PATH to include the proper directory
+// for the dynamic linker
+if (process.platform === 'win32') {
+    const dslib_path = path.resolve(path.join(binding_path, '../..'));
+    var oldPath = process.env.PATH;
+    process.env['PATH'] = `${dslib_path};${process.env.PATH}`;
+}
+
+const binding = require(binding_path);
+
+if (process.platform === 'win32') {
+  process.env['PATH'] = oldPath;
+}
+
+/**
+ * Stores text of an individual token, along with its timing information
+ */
+export interface TokenMetadata {
+    /** The text corresponding to this token */
+    text: string;
+
+    /** Position of the token in units of 20ms */
+    timestep: number;
+
+    /** Position of the token in seconds */
+    start_time: number;
+}
+
+/**
+ * A single transcript computed by the model, including a confidence value and
+ * the metadata for its constituent tokens.
+ */
+export interface CandidateTranscript {
+    tokens: TokenMetadata[];
+
+    /**
+     * Approximated confidence value for this transcription. This is roughly the
+     * sum of the acoustic model logit values for each timestep/token that
+     * contributed to the creation of this transcription.
+     */
+    confidence: number;
+}
+
+/**
+ * An array of CandidateTranscript objects computed by the model.
+ */
+export interface Metadata {
+    transcripts: CandidateTranscript[];
+}
+
+/**
+ * Provides an interface to a DeepSpeech stream. The constructor cannot be called
+ * directly, use :js:func:`Model.createStream`.
+ */
+class Stream {
+    /** @internal */
+    _impl: any;
+
+    /**
+     * @param nativeStream SWIG wrapper for native StreamingState object.
+     */
+    constructor(nativeStream: object) {
+        this._impl = nativeStream;
+    }
+
+    /**
+     * Feed audio samples to an ongoing streaming inference.
+     *
+     * @param aBuffer An array of 16-bit, mono raw audio samples at the
+     *                 appropriate sample rate (matching what the model was trained on).
+     */
+    feedAudioContent(aBuffer: Buffer): void {
+        binding.FeedAudioContent(this._impl, aBuffer);
+    }
+
+    /**
+     * Compute the intermediate decoding of an ongoing streaming inference.
+     *
+     * @return The STT intermediate result.
+     */
+    intermediateDecode(): string {
+        return binding.IntermediateDecode(this._impl);
+    }
+
+    /**
+     * Compute the intermediate decoding of an ongoing streaming inference, return results including metadata.
+     *
+     * @param aNumResults Maximum number of candidate transcripts to return. Returned list might be smaller than this. Default value is 1 if not specified.
+     *
+     * @return :js:func:`Metadata` object containing multiple candidate transcripts. Each transcript has per-token metadata including timing information. The user is responsible for freeing Metadata by calling :js:func:`FreeMetadata`. Returns undefined on error.
+     */
+    intermediateDecodeWithMetadata(aNumResults: number = 1): Metadata {
+        return binding.IntermediateDecodeWithMetadata(this._impl, aNumResults);
+    }
+
+    /**
+     * Compute the final decoding of an ongoing streaming inference and return the result. Signals the end of an ongoing streaming inference.
+     *
+     * @return The STT result.
+     *
+     * This method will free the stream, it must not be used after this method is called.
+     */
+    finishStream(): string {
+        const result = binding.FinishStream(this._impl);
+        this._impl = null;
+        return result;
+    }
+
+    /**
+     * Compute the final decoding of an ongoing streaming inference and return the results including metadata. Signals the end of an ongoing streaming inference.
+     *
+     * @param aNumResults Maximum number of candidate transcripts to return. Returned list might be smaller than this. Default value is 1 if not specified.
+     *
+     * @return Outputs a :js:func:`Metadata` struct of individual letters along with their timing information. The user is responsible for freeing Metadata by calling :js:func:`FreeMetadata`.
+     *
+     * This method will free the stream, it must not be used after this method is called.
+     */
+    finishStreamWithMetadata(aNumResults: number = 1): Metadata {
+        const result = binding.FinishStreamWithMetadata(this._impl, aNumResults);
+        this._impl = null;
+        return result;
+    }
+}
+
+/**
+ * An object providing an interface to a trained DeepSpeech model.
+ */
+export class Model {
+    /** @internal */
+    _impl: any;
+
+    /**
+     * @param aModelPath The path to the frozen model graph.
+     *
+     * @throws on error
+     */
+    constructor(aModelPath: string) {
+        this._impl = null;
+
+        const [status, impl] = binding.CreateModel(aModelPath);
+        if (status !== 0) {
+            throw `CreateModel failed: ${binding.ErrorCodeToErrorMessage(status)} (0x${status.toString(16)})`;
+        }
+
+        this._impl = impl;
+    }
+
+    /**
+     * Get beam width value used by the model. If :js:func:`Model.setBeamWidth` was
+     * not called before, will return the default value loaded from the model file.
+     *
+     * @return Beam width value used by the model.
+     */
+    beamWidth(): number {
+        return binding.GetModelBeamWidth(this._impl);
+    }
+
+    /**
+     * Set beam width value used by the model.
+     *
+     * @param aBeamWidth The beam width used by the model. A larger beam width value generates better results at the cost of decoding time.
+     *
+     * @throws on error
+     */
+    setBeamWidth(aBeamWidth: number): void {
+        const status = binding.SetModelBeamWidth(this._impl, aBeamWidth);
+        if (status !== 0) {
+            throw `SetModelBeamWidth failed: ${binding.ErrorCodeToErrorMessage(status)} (0x${status.toString(16)})`;
+        }
+    }
+
+    /**
+     * Return the sample rate expected by the model.
+     *
+     * @return Sample rate.
+     */
+    sampleRate(): number {
+        return binding.GetModelSampleRate(this._impl);
+    }
+
+    /**
+     * Enable decoding using an external scorer.
+     *
+     * @param aScorerPath The path to the external scorer file.
+     *
+     * @throws on error
+     */
+    enableExternalScorer(aScorerPath: string): void {
+        const status = binding.EnableExternalScorer(this._impl, aScorerPath);
+        if (status !== 0) {
+            throw `EnableExternalScorer failed: ${binding.ErrorCodeToErrorMessage(status)} (0x${status.toString(16)})`;
+        }
+    }
+
+    /**
+     * Disable decoding using an external scorer.
+     *
+     * @throws on error
+     */
+    disableExternalScorer(): void {
+        const status = binding.DisableExternalScorer(this._impl);
+        if (status !== 0) {
+            throw `DisableExternalScorer failed: ${binding.ErrorCodeToErrorMessage(status)} (0x${status.toString(16)})`;
+        }
+    }
+
+    /**
+     * Set hyperparameters alpha and beta of the external scorer.
+     *
+     * @param aLMAlpha The alpha hyperparameter of the CTC decoder. Language Model weight.
+     * @param aLMBeta The beta hyperparameter of the CTC decoder. Word insertion weight.
+     *
+     * @throws on error
+     */
+    setScorerAlphaBeta(aLMAlpha: number, aLMBeta: number): void {
+        const status = binding.SetScorerAlphaBeta(this._impl, aLMAlpha, aLMBeta);
+        if (status !== 0) {
+            throw `SetScorerAlphaBeta failed: ${binding.ErrorCodeToErrorMessage(status)} (0x${status.toString(16)})`;
+        }
+    }
+
+    /**
+     * Use the DeepSpeech model to perform Speech-To-Text.
+     *
+     * @param aBuffer A 16-bit, mono raw audio signal at the appropriate sample rate (matching what the model was trained on).
+     *
+     * @return The STT result. Returns undefined on error.
+     */
+    stt(aBuffer: Buffer): string {
+        return binding.SpeechToText(this._impl, aBuffer);
+    }
+
+    /**
+     * Use the DeepSpeech model to perform Speech-To-Text and output metadata
+     * about the results.
+     *
+     * @param aBuffer A 16-bit, mono raw audio signal at the appropriate sample rate (matching what the model was trained on).
+     * @param aNumResults Maximum number of candidate transcripts to return. Returned list might be smaller than this.
+     * Default value is 1 if not specified.
+     *
+     * @return :js:func:`Metadata` object containing multiple candidate transcripts. Each transcript has per-token metadata including timing information.
+     * The user is responsible for freeing Metadata by calling :js:func:`FreeMetadata`. Returns undefined on error.
+     */
+    sttWithMetadata(aBuffer: Buffer, aNumResults: number = 1): Metadata {
+        return binding.SpeechToTextWithMetadata(this._impl, aBuffer, aNumResults);
+    }
+
+    /**
+     * Create a new streaming inference state. One can then call :js:func:`Stream.feedAudioContent` and :js:func:`Stream.finishStream` on the returned stream object.
+     *
+     * @return a :js:func:`Stream` object that represents the streaming state.
+     *
+     * @throws on error
+     */
+    createStream(): Stream {
+        const [status, ctx] = binding.CreateStream(this._impl);
+        if (status !== 0) {
+            throw `CreateStream failed: ${binding.ErrorCodeToErrorMessage(status)} (0x${status.toString(16)})`;
+        }
+        return new Stream(ctx);
+    }
+}
+
+/**
+ * Frees associated resources and destroys model object.
+ *
+ * @param model A model pointer returned by :js:func:`Model`
+ *
+ */
+export function FreeModel(model: Model): void {
+    binding.FreeModel(model._impl);
+}
+
+/**
+ * Free memory allocated for metadata information.
+ *
+ * @param metadata Object containing metadata as returned by :js:func:`Model.sttWithMetadata` or :js:func:`Stream.finishStreamWithMetadata`
+ */
+export function FreeMetadata(metadata: Metadata): void {
+    binding.FreeMetadata(metadata);
+}
+
+/**
+ * Destroy a streaming state without decoding the computed logits. This
+ * can be used if you no longer need the result of an ongoing streaming
+ * inference and don't want to perform a costly decode operation.
+ *
+ * @param stream A streaming state pointer returned by :js:func:`Model.createStream`.
+ */
+export function FreeStream(stream: Stream): void {
+    binding.FreeStream(stream._impl);
+}
+
+/**
+ * Returns the version of this library. The returned version is a semantic
+ * version (SemVer 2.0.0).
+ */
+export function Version(): string {
+    return binding.Version();
+}
diff --git a/native_client/javascript/node-pre-gyp.d.ts b/native_client/javascript/node-pre-gyp.d.ts
new file mode 100644
index 00000000..0bb1cfb5
--- /dev/null
+++ b/native_client/javascript/node-pre-gyp.d.ts
@@ -0,0 +1,3 @@
+declare module "node-pre-gyp" {
+    export function find(path: String): any;
+}
diff --git a/native_client/javascript/package.json.in b/native_client/javascript/package.json.in
index 22c3363c..00079851 100644
--- a/native_client/javascript/package.json.in
+++ b/native_client/javascript/package.json.in
@@ -41,7 +41,8 @@
     "devDependencies": {
       "electron": "^1.7.9",
       "node-gyp": "4.x - 5.x",
-      "typescript": "3.6.x",
+      "typescript": "3.8.x",
+      "typedoc": "0.17.x",
       "@types/argparse": "1.0.x",
       "@types/node": "13.9.x"
     },
diff --git a/native_client/javascript/tsconfig.json b/native_client/javascript/tsconfig.json
index d67d7bce..05021449 100644
--- a/native_client/javascript/tsconfig.json
+++ b/native_client/javascript/tsconfig.json
@@ -9,10 +9,14 @@
         "noImplicitThis": true,
         "strictFunctionTypes": true,
         "strictNullChecks": true,
-        "forceConsistentCasingInFileNames": true
+        "forceConsistentCasingInFileNames": true,
+        "alwaysStrict": true,
+        "declaration": true,
+        "stripInternal": true
     },
     "files": [
-        "index.d.ts",
-        "client.ts"
+        "index.ts",
+        "client.ts",
+        "node-pre-gyp.d.ts"
     ]
 }
diff --git a/taskcluster/docs-requirements.txt b/taskcluster/docs-requirements.txt
index 4530be26..7dc4b281 100644
--- a/taskcluster/docs-requirements.txt
+++ b/taskcluster/docs-requirements.txt
@@ -1,6 +1,7 @@
 breathe==4.14.2
 semver==2.8.1
 sphinx==2.4.4
-sphinx-js==2.8
+#FIXME: switch back to upstream sphinx-js when https://github.com/mozilla/sphinx-js/pull/135 is merged or the issue is fixed otherwise
+git+git://github.com/reuben/sphinx-js.git@a24775935443d21028ee4a7025a407c78030c4e7#egg=sphinx-js
 sphinx-rtd-theme==0.4.3
 pygments==2.6.1